[ { "id": "qwen", "name": "Qwen", "company": "Alibaba", "region": "china", "tagline": "Alibaba foundation model and AI assistant family", "description": "Qwen is Alibaba Cloud's foundation-model family and chat surface. The catalog treats Qwen as the canonical suite-level row for Qwen Chat, Qwen model APIs, coding, image, video, deep-research, PPT, and artifact workflows rather than splitting each mode into separate products.", "users": "Not disclosed", "userCount": 0, "releaseDate": "2023-08", "technicalImpact": 90, "competitors": [ "deepseek", "llama", "gemini", "claude", "grok" ], "features": [ "Language, code, vision, image and audio models", "Open-weight and commercial deployments", "Alibaba Cloud and Qwen app distribution", "Model Studio integration", "Frequent generation cadence across modalities" ], "pricing": { "open_source": "Free for open-weight", "api": "Alibaba Cloud pricing" }, "link": "https://qwen.ai/", "apiLink": "https://bailian.console.aliyun.com/", "aiCoreShare": 100, "adoptionSignal": "Qwen is reported to have reached 300M monthly active users across Alibaba's consumer platforms by early 2026, and Alipay processed 120M AI-agent transactions in a single week in February 2026. Both figures are bundled across Alibaba surfaces rather than Qwen-specific adoption, so they are recorded qualitatively and `userCount` stays 0. Qwen is also the most-downloaded model family on ModelScope.", "sourceUrls": [ "https://qwen.ai/", "https://www.alibabacloud.com/product/modelstudio", "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "https://www.alibabacloud.com/help/en/model-studio/models" ], "nativeName": "通义千问", "scoringRationale": "Qwen is Alibaba Cloud's frontier model family (Qwen2.5, Qwen3, multimodal) with strong open-weight benchmark performance and Alibaba Cloud API deployment. Scored at 90 to reflect frontier-model quality and cloud-platform distribution without disclosed product-specific user counts — placing it at open-weight frontier tier, below ChatGPT/Claude/Gemini consumer products with confirmed massive adoption, and consistent with other frontier open-weight families at this score tier.", "versions": [ { "name": "Qwen 3.8", "released": "2026-08", "variants": [ "Qwen3.8 Max", "Qwen3.8 Flash", "1M-token context", "Native visual understanding" ], "note": "Current multimodal generation with Max released August 2 and Flash released August 26." }, { "name": "Qwen 3.7 Max" }, { "name": "Qwen 3.5" }, { "name": "Qwen3", "variants": [ "Coder", "VL", "dense and MoE sizes" ] } ], "versionsUpdated": "2026-09-02" }, { "id": "ernie", "name": "ERNIE", "company": "Baidu", "region": "china", "tagline": "Baidu's ERNIE model and assistant family", "description": "ERNIE is Baidu's foundation-model and assistant family, distributed to developers through Baidu Qianfan and to consumers through ERNIE Bot. Versioned model and assistant rows resolve to this canonical family page.", "users": "Not disclosed", "userCount": 0, "releaseDate": "2023-10", "technicalImpact": 80, "competitors": [ "qwen", "deepseek", "hyperclova-x", "GPT-5", "claude", "Gemini 2.5", "Grok 4", "chatgpt", "doubao" ], "features": [ "Foundation model family", "Reasoning and assistant variants", "Baidu AI Cloud integration", "Consumer and enterprise access", "Suite modules: ERNIE Bot", "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Baidu Cloud pricing", "enterprise": "Enterprise agreements" }, "link": "https://cloud.baidu.com/product-s/qianfan_home", "apiLink": "https://cloud.baidu.com/product-s/qianfan_home", "aiCoreShare": 100, "adoptionSignal": "Qianfan publishes model and platform surfaces, but the catalog does not record a current ERNIE-product user count.", "sourceUrls": [ "https://cloud.baidu.com/product-s/qianfan_home", "https://ir.baidu.com/static-files/cd96c440-1fbc-4346-9d47-4248b2ef77c0", "https://yiyan.baidu.com/" ], "nativeName": "文心 5.0", "scoringRationale": "Baidu ERNIE (5.0 and successors) is China's longest-running LLM product with reported 200M+ registered users on ERNIE Bot across prior versions. ERNIE 5.0 specifically lacks disclosed deployment volume and benchmark evidence in the catalog. Scored at 80 rather than 84 because ERNIE-specific production evidence for the current version is not disclosed and Baidu's AI metrics mix search, cloud, and assistant surfaces.", "versions": [ { "name": "ERNIE 5.1", "note": "Current flagship surfaced through Baidu Qianfan." }, { "name": "ERNIE 5.0", "released": "2026-01" }, { "name": "ERNIE 4.5" } ], "versionsUpdated": "2026-09-02" }, { "id": "yi-foundation-models", "name": "Yi Foundation Models", "company": "01.AI", "region": "china", "tagline": "01.AI open and commercial foundation-model family", "description": "Yi Foundation Models is 01.AI's canonical model-family row for Yi-Lightning, Yi-Large, Yi-VL, Yi-1.5, and Yi-Coder. The official product page positions Yi-Lightning and Yi-Large as flagship LLMs, lists open-source distribution through Hugging Face, ModelScope, GitHub, and a technical report, and describes Yi-VL multimodal models plus Yi-Coder's 128K-context code generation across 52 programming languages. Public product-specific active users, enterprise customers, and revenue are not disclosed, so the score stays below DeepSeek, Qwen, and stronger disclosed-adoption assistants.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "01.AI's Yi page lists Yi-Lightning, Yi-Large, Yi-VL, Yi-1.5 and Yi-Coder with Hugging Face, ModelScope and GitHub distribution. 01.AI has since pivoted toward DeepSeek-based models, leaving Yi as a legacy family; current usage and deployment are not disclosed.", "releaseDate": "2023-11", "technicalImpact": 80, "competitors": [ "qwen", "deepseek", "ernie", "kimi", "glm-4" ], "features": [ "Yi-Lightning and Yi-Large model family", "Yi-VL multimodal models", "Yi-1.5 chat and base models", "Yi-Coder 1.5B and 9B code models", "128K code context", "Open-source distribution paths", "Chinese and English model coverage", "Commercial license path" ], "pricing": { "open_source": "Free weights where available", "commercial": "Apply for commercial license", "enterprise": "Contact 01.AI" }, "link": "https://www.01.ai/yi-models", "apiLink": "https://www.01.ai/yi-models", "aiCoreShare": 100, "sourceUrls": [ "https://www.01.ai/", "https://www.01.ai/yi-models", "https://platform.01.ai/docs", "https://arxiv.org/abs/2403.04652" ], "scoringRationale": "Reduced from 84 to 80 on cross-region fairness. 01.AI pivoted to enterprise deployments after Yi-Lightning, so 84 overstated the family against actively advancing global-ecosystem peers. Yi keeps credit for open-weight distribution and benchmark credibility; current product-specific usage remains undisclosed.", "versions": [ { "name": "Yi-Lightning", "released": "2024-10", "variants": [ "Yi-Large", "Yi-VL", "Yi-Coder" ], "note": "Latest first-party Yi generation listed by 01.AI and its API documentation." }, { "name": "Yi 1.5" } ], "versionsUpdated": "2026-09-02" }, { "id": "deepseek", "name": "DeepSeek", "company": "DeepSeek", "region": "china", "tagline": "Open-weight frontier reasoning and agent model family", "description": "DeepSeek V4 is DeepSeek's current flagship model family, advertised for frontier reasoning, stronger agent behavior, web/app/API access, and open-model research continuity. Older DeepSeek V2 and V3 rows should resolve conceptually into this canonical DeepSeek surface.", "users": "Not disclosed", "userCount": 0, "releaseDate": "2024-12", "technicalImpact": 90, "competitors": [ "llama", "qwen", "gpt-5-5", "gemini", "GPT-5", "claude", "Gemini 2.5", "Grok 4", "github-copilot", "cursor" ], "features": [ "Open-weight frontier models", "Reasoning and coding models", "Multimodal model variants", "Consumer app and API access", "Suite modules: DeepSeek API, DeepSeek Coder, DeepSeek Coder V2, DeepSeek Janus-Pro, DeepSeek Math, DeepSeek Prover, DeepSeek R1, DeepSeek R1-Zero, DeepSeek V3.1, DeepSeek VL2", "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization", "Open-source model", "Code completion", "Multi-language", "MIT license" ], "pricing": { "api": "Peak/off-peak usage pricing", "open_source": "Free weights" }, "link": "https://www.deepseek.com/", "apiLink": "https://api-docs.deepseek.com/", "aiCoreShare": 100, "adoptionSignal": "DeepSeek publishes the app/API and model releases, but the catalog does not record a current active-user count.", "sourceUrls": [ "https://www.deepseek.com/", "https://api-docs.deepseek.com/news/news260813/", "https://api-docs.deepseek.com/updates" ], "nativeName": "深度求索", "scoringRationale": "DeepSeek is a frontier open-weight model family (V3, R1, and successors) that benchmarks competitively with closed frontier models on coding and reasoning tasks. Scored at 90 to reflect genuine frontier-model quality and open-weight ecosystem adoption without disclosed commercial user counts or revenue — placing it below ChatGPT/Claude/Gemini consumer products (which have hundreds of millions of users) but above Gemma (90, comparable open-weight evidence) due to stronger reasoning benchmark differentiation.", "versions": [ { "name": "DeepSeek V4 Pro", "released": "2026-08", "variants": [ "V4 Pro", "V4 Flash", "V4 Flash Vision Experimental", "Flexible reasoning effort" ], "note": "General-availability V4 generation with native Responses API support and app/web Expert Mode." }, { "name": "DeepSeek V3.1" }, { "name": "DeepSeek R1", "note": "Open-weight reasoning release that established the family internationally." } ], "versionsUpdated": "2026-09-02" }, { "id": "llama", "name": "Llama", "company": "Meta", "region": "global", "tagline": "Meta open-weight models for existing and self-hosted deployments", "description": "Llama is Meta's open-weight model family for self-hosted and cloud deployment. Released Llama 4 weights include Scout and Maverick. Existing integrations and fine-tunes remain useful, but historical downloads do not establish current leadership or developer preference. Llama uses a custom community licence, not Apache 2.0.", "users": "Current active deployment count not disclosed", "userCount": 0, "releaseDate": "2025-04", "technicalImpact": 80, "competitors": [ "gemma", "qwen", "deepseek", "mistral" ], "features": [ "Open-weight language and multimodal models", "Scout and Maverick model weights", "Self-hosting and fine-tuning", "Custom Llama community licence" ], "pricing": { "open_source": "Free", "cloud": "AWS, Azure, GCP" }, "link": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "apiLink": null, "aiCoreShare": 100, "sourceUrls": [ "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct" ], "scoringRationale": "Reviewed 2026-09-29: Impact 84 → 80. Removes unsupported present-tense claims of the largest or dominant open-weight ecosystem. Established integrations retain value; historical downloads cannot prove current preference over Gemma, Qwen or DeepSeek.", "versions": [ { "name": "Llama 4", "released": "2025-04", "variants": [ "Scout", "Maverick" ], "note": "Released weight variants verified; Behemoth is not presented as an available download." }, { "name": "Llama 3.3 / 3.2 / 3.1", "note": "Earlier generations remain relevant to existing deployments." } ], "versionsUpdated": "2026-09-29", "adoptionSignal": "Released weights and deployment guidance remain available. No current active-deployment count or comparative market-share measurement was verified." }, { "id": "hyperclova-x", "name": "HyperCLOVA X", "company": "Naver", "region": "korea", "tagline": "Korea's sovereign AI foundation model", "description": "HyperCLOVA X is Naver's Korean foundation-model family, now centered on the SEED lineup: SEED 32B Think (reasoning), SEED 8B Omni (any-to-any omnimodal), and the defense-focused SEED 4B unveiled June 2026 for real-time inference on drones and tactical vehicles. Naver Cloud was dropped from Korea's sovereign-AI foundation-model project in the January 2026 phase-1 cut and has since aligned with NVIDIA (AI-factory alliance; plans to fine-tune Nemotron with proprietary data). Disclosed model-API adoption remains limited.", "descriptionKo": "HyperCLOVA X는 네이버가 개발한 한국어 파운데이션 모델 패밀리이자 소버린 AI 플랫폼입니다. 2026년 기준으로 오픈웨이트 한국어 추론 모델과 멀티모달(옴니모달) 기술 릴리스를 확인할 수 있으며, HyperCLOVA X 32B Think(추론), 8B Omni(비전-언어), 32B 비전-언어 모델 등이 포함됩니다. 한국어 문화·언어 이해와 멀티링궐 지원에 강점이 있고, 기업용 배포도 지원합니다. 다만 모델 API 유료 고객 수나 외부 배포 규모는 공개되지 않아, Llama·Qwen 같은 오픈웨이트 글로벌 모델 대비 배포 증거가 부족한 상태입니다.", "users": "Not disclosed", "userCount": 0, "releaseDate": "2023-08", "technicalImpact": 80, "competitors": [ "exaone", "solar-llm", "kanana", "qwen", "gemini" ], "features": [ "82B+ parameters", "Korean cultural understanding", "Multilingual", "Enterprise deployment", "Suite modules: HyperCLOVA X 32B Think, HyperCLOVA X 8B Omni", "32B vision-language model", "Korean linguistic and cultural reasoning", "Agentic behavior support", "Open-weight release", "Text, audio, and vision inputs", "Text, audio, and vision outputs", "8B-scale omnimodal architecture" ], "pricing": { "api": "Contact Naver Cloud", "enterprise": "Custom agreements" }, "link": "https://clova.ai/hyperclova", "apiLink": "https://www.ncloud.com/product/aiService/clovaStudio", "aiCoreShare": 100, "nativeName": "하이퍼클로바X", "sourceUrls": [ "https://arxiv.org/abs/2601.03286", "https://arxiv.org/abs/2601.01792", "https://www.navercorp.com/en/media/pressReleasesDetail?seq=34256", "https://en.sedaily.com/news/2026/06/15/naver-unveils-defense-omnimodal-ai-for-real-time-drone" ], "scoringRationale": "Reduced from 84 to 80 in the 2026-07-29 fairness pass, applying the adoption rule uniformly: high-spend models with weak disclosed usage do not score top-tier on benchmarks alone. Korea-leading Artificial Analysis standing, Naver distribution, a defense contract and the sovereign mandate are real and sourced — that earns the 80 tier above domestic model rows without third-party recognition (78). But 84 is the tier of adoption evidence (Llama, 500M+ downloads) or global developer-ecosystem roles (GLM, Kimi), and HyperCLOVA X's standing is regional with no disclosed product-specific adoption. Naver-platform reach is parent-platform scale and is not creditable as product adoption.", "adoptionSignal": "Naver Cloud's HyperCLOVA X SEED Think 32B was reported as the leading Korean model on the Artificial Analysis Intelligence Index in early 2026 — third-party benchmark standing rather than a vendor claim. Naver Cloud is one of five consortia in Korea's national sovereign foundation-model programme. Product-specific user counts are not disclosed. Korea's Ministry of Science and ICT selected five consortia — LG AI Research, SK Telecom, Naver Cloud, NC AI and Upstage — for the national sovereign foundation-model mandate under a ~$390M programme reviewed every six months. MSIT opened bidding on 14 July 2026 for \"AI for Everyone\", a free national assistant required to run at least 50% on domestic models, with beta targeted for late September 2026.", "versions": [ { "name": "HyperCLOVA X SEED Think 32B", "released": "2026", "variants": [ "SEED Think 32B", "SEED lineup" ], "note": "Reported as the leading Korean model on the Artificial Analysis Intelligence Index in early 2026." }, { "name": "HyperCLOVA X Think", "released": "2025-06" }, { "name": "HyperCLOVA X", "released": "2023-08" } ], "versionsUpdated": "2026-07-29" }, { "id": "gpt-oss", "name": "gpt-oss", "company": "OpenAI", "region": "global", "tagline": "Open-weight reasoning models from OpenAI.", "description": "gpt-oss is OpenAI's open-weight reasoning model family, available in 120B and 20B variants for self-hosted agentic, coding, and reasoning workloads.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "OpenAI says gpt-oss weights are free to download under Apache 2.0 and run on self-managed infrastructure or hosting partners; active deployment counts are not disclosed.", "releaseDate": "2025-08", "technicalImpact": 84, "competitors": [ "llama", "qwen", "deepseek", "gemma-4" ], "features": [ "Open-weight reasoning models", "120B and 20B variants", "Apache 2.0 license", "Tool use and structured outputs", "Self-hosted deployment" ], "pricing": { "open_weight": "Free weights", "hosting": "Self-hosted or third-party hosting costs" }, "link": "https://openai.com/open-models/", "apiLink": "https://openai.com/index/introducing-gpt-oss/", "aiCoreShare": 100, "sourceUrls": [ "https://openai.com/open-models/", "https://openai.com/index/introducing-gpt-oss/", "https://help.openai.com/en/articles/11870455-openai-open-weight-models-gpt-oss" ], "scoringRationale": "OpenAI's open-weight reasoning model family (120B + 20B variants) for self-hosted agentic and coding workloads. Strong frontier-model credentials from the OpenAI brand and GPQA/coding benchmark scores, but scored at 84 rather than higher because deployment volume, developer download counts, and ecosystem adoption are not yet disclosed — the model launched recently with no public usage evidence in the catalog.", "versions": [ { "name": "gpt-oss", "released": "2025-08", "variants": [ "120B", "20B" ], "note": "Apache 2.0 open weights for self-hosted reasoning and agentic workloads." } ], "versionsUpdated": "2026-07-29" }, { "id": "askup", "name": "AskUp", "company": "Upstage", "region": "korea", "tagline": "Upstage's messenger-based AI assistant with document OCR", "description": "AskUp is Upstage's consumer AI assistant delivered through KakaoTalk in Korea and LINE in Japan. It combines conversational search with Upstage OCR for extracting and summarizing text from images; Upstage does not disclose a current AskUp-specific active-user count.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 77, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.upstage.ai/apps/askup", "apiLink": "https://console.upstage.ai", "aiCoreShare": 100, "nativeName": "아숙업", "sourceUrls": [ "https://www.upstage.ai/apps/askup" ], "adoptionSignal": "Current first-party pages confirm availability through KakaoTalk and LINE, plus OCR and Google Search integration; current product-specific usage is not disclosed.", "versions": [ { "name": "AskUp", "variants": [ "KakaoTalk", "LINE" ] } ], "versionsUpdated": "2026-09-02" }, { "id": "txgemma", "name": "TxGemma", "company": "Google", "region": "global", "tagline": "Therapeutics and drug-discovery open model family from Google", "description": "TxGemma is Google's open-weight model family tuned for therapeutic development — property prediction, molecule reasoning and drug-discovery tasks — derived from Gemma for self-hosted research use. Google discloses no deployment figures.", "users": "Not disclosed", "userCount": 0, "releaseDate": "2026-04", "technicalImpact": 76, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://ai.google.dev", "apiLink": "https://ai.google.dev/gemini-api/docs/models", "aiCoreShare": 100, "versions": [ { "name": "TxGemma", "released": "2026-04", "note": "Gemma-derived; therapeutics and drug-discovery tasks." } ], "versionsUpdated": "2026-07-29" }, { "id": "naver-ai-briefing", "name": "Naver AI Briefing", "company": "Naver", "region": "korea", "tagline": "AI-generated search briefings embedded in Naver Search", "description": "Naver AI Briefing generates source-linked summaries inside Naver Search across general queries, places, short-form content and shopping. Naver launched it broadly in March 2025 and describes a potential 40 million-user search audience, but that parent-search exposure is not counted as product adoption.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 74, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.navercorp.com/media/pressReleasesDetail?seq=32400", "apiLink": "https://www.ncloud.com/product/aiService/clovaStudio", "aiCoreShare": 100, "scoringRationale": "Held at 74 under the bundled-exposure rule: Naver says AI Briefing can reach its 40 million-user search audience, but does not disclose AI Briefing-specific usage, so parent-search scale cannot carry Impact.", "adoptionSignal": "Launched to all Naver Search users in March 2025; the stated 40 million-user audience is bundled search exposure, not disclosed AI Briefing adoption.", "sourceUrls": [ "https://www.navercorp.com/media/pressReleasesDetail?seq=32400", "https://www.navercorp.com/en/service/featuredServices" ], "versions": [ { "name": "AI Briefing", "released": "2025-03", "variants": [ "Search", "Places", "Short-form", "Shopping" ] } ], "versionsUpdated": "2026-09-02" }, { "id": "nugu", "name": "NUGU", "company": "SK Telecom", "region": "korea", "tagline": "SK Telecom's Korean voice-assistant and device ecosystem", "description": "NUGU is SK Telecom's Korean voice-assistant platform across speakers, connected devices and supported service integrations. The service and developer documentation remain available, but SK Telecom does not disclose a current NUGU-specific active-user count.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.nugu.co.kr/", "apiLink": null, "aiCoreShare": 100, "nativeName": "누구", "sourceUrls": [ "https://www.nugu.co.kr/" ], "adoptionSignal": "Current service manuals and support pages confirm the NUGU device and integration surface; current active users are not disclosed.", "versions": [ { "name": "NUGU", "variants": [ "NUGU devices", "T Phone integration", "Developer services" ] } ], "versionsUpdated": "2026-09-02" }, { "id": "yuanbao", "name": "Yuanbao", "company": "Tencent", "region": "china", "tagline": "Tencent AI assistant", "description": "Yuanbao is Tencent's AI assistant app, based on Hunyuan and other supported models. The catalog treats it as Tencent's consumer assistant surface rather than another Hunyuan model row.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 80, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://yuanbao.tencent.com/", "apiLink": null, "aiCoreShare": 100, "nativeName": "元宝", "sourceUrls": [ "https://yuanbao.tencent.com/" ], "scoringRationale": "Tencent Yuanbao is the consumer assistant surface for Hunyuan-backed workflows. Tencent distribution gives it weight, but product-specific active users are not disclosed.", "versions": [ { "name": "Yuanbao", "note": "Tencent's consumer assistant, running on Hunyuan and other supported models." } ], "versionsUpdated": "2026-07-29" }, { "id": "capcut-ai", "name": "CapCut AI", "company": "ByteDance", "region": "china", "tagline": "AI video creation and editing tools in CapCut", "description": "CapCut AI covers AI-assisted video generation and editing workflows inside CapCut. The direct product link now points to a CapCut AI tool rather than the generic Doubao or ByteDance homepage.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.capcut.com/tools/ai-video", "apiLink": null, "aiCoreShare": 100, "sourceUrls": [ "https://www.capcut.com/tools/ai-video", "https://www.capcut.com/tools" ], "versions": [ { "name": "CapCut AI Video Maker 2.0", "variants": [ "Seedance 2.5", "Veo 3.1", "Sora 2" ], "note": "Current CapCut AI video creation surface and integrated model options." } ], "versionsUpdated": "2026-09-02" }, { "id": "gpt-5-5", "name": "GPT", "company": "OpenAI", "region": "global", "tagline": "OpenAI's frontier GPT model family, behind ChatGPT and the OpenAI API", "description": "GPT is OpenAI's model family for reasoning, coding and multimodal applications. The recorded lineup includes GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. ChatGPT and Codex are separate products built on these models. Use Responses for agent integrations and check model-specific endpoint and tool support before migration. Model-family user counts are not disclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "2026-05", "technicalImpact": 90, "competitors": [ "claude", "gemini", "grok", "qwen", "deepseek" ], "features": [ "Text and image input", "Reasoning and coding", "Responses API", "Model-specific tool and endpoint support" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://developers.openai.com/api/docs/models", "apiLink": "https://developers.openai.com/api/docs/models", "aiCoreShare": 100, "sourceUrls": [ "https://platform.openai.com/docs/models", "https://openai.com/api/pricing/", "https://developers.openai.com/api/docs/models", "https://developers.openai.com/api/docs/models/gpt-6.1-sol" ], "versions": [ { "name": "GPT-6 family", "variants": [ "GPT-6.1 Sol", "GPT-6 Astra", "GPT-6 Luna", "GPT-6 Sol (earlier Sol release)" ], "note": "Lineup checked against the official model catalog; API support differs by model." }, { "name": "GPT-5.6", "variants": [ "Sol", "Terra", "Luna" ], "note": "Earlier generation." }, { "name": "GPT-5.5" }, { "name": "GPT-5" }, { "name": "o3 / o4-mini", "note": "Earlier reasoning models; check deprecation notices before use." } ], "versionsUpdated": "2026-10-01", "scoringRationale": "Impact 90 reflects first-party frontier-model control, commercial API distribution and its role behind ChatGPT and Codex. App audience figures are not attributed to this API model-family row. This version refresh does not establish a new adoption count or increase the score.", "descriptionKo": "GPT는 추론·코딩·멀티모달 애플리케이션을 위한 OpenAI 모델 제품군입니다. 확인된 제품군은 GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna입니다. ChatGPT와 Codex는 별도 사용자용 제품입니다. 에이전트 연동은 Responses를 사용하고 모델별 엔드포인트·도구 지원을 확인하세요. 모델 제품군의 사용자 수는 공개되지 않았습니다." }, { "id": "tongyi-wanxiang", "name": "Wan", "company": "Alibaba", "region": "china", "tagline": "Alibaba's Wan image and video generation family", "description": "Wan is Alibaba's creative-generation family for image creation, editing and video workflows through Model Studio. The older Tongyi Wanxiang name and route now resolve to the current Wan family rather than representing a separate product.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.alibabacloud.com/help/en/model-studio/use-video-generation/", "apiLink": "https://www.alibabacloud.com/help/en/model-studio/wan-image-generation-and-editing-api-reference", "aiCoreShare": 100, "sourceUrls": [ "https://www.alibabacloud.com/help/en/model-studio/use-video-generation/", "https://www.alibabacloud.com/help/en/model-studio/wan-image-generation-and-editing-api-reference" ], "versions": [ { "name": "Wan 3.0 Video", "released": "2026-08" }, { "name": "Wan 2.7", "variants": [ "Image", "Image Pro", "Video" ] } ], "versionsUpdated": "2026-09-02" }, { "id": "tencent-codebuddy", "name": "Tencent CodeBuddy", "company": "Tencent", "region": "china", "tagline": "Tencent Cloud AI coding assistant and IDE", "description": "Tencent CodeBuddy is Tencent Cloud's coding assistant and AI IDE surface, using Hunyuan and DeepSeek-backed workflows for product, design, development, and deployment tasks.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 80, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.codebuddy.ai/", "apiLink": null, "aiCoreShare": 100, "sourceUrls": [ "https://www.codebuddy.ai/", "https://www.codebuddy.ai/docs/ide/release-notes/release-notes" ], "scoringRationale": "Tencent CodeBuddy is a maintained AI coding and IDE surface, but remains below Claude Code, Cursor and GitHub Copilot because public developer adoption is not disclosed.", "versions": [ { "name": "CodeBuddy IDE 4.10.4", "released": "2026-07", "note": "Latest documented IDE release at the September 2026 review." } ], "versionsUpdated": "2026-09-02" }, { "id": "solar-llm", "name": "Solar LLM", "company": "Upstage", "region": "korea", "tagline": "Efficient Korean enterprise LLM", "description": "Solar is Upstage's Korean enterprise LLM family. Solar Pro 3 (January 2026, 128K context) is the commercial flagship, and Solar Open 100B is the open sovereign-track model; Solar Pro 2 was the first Korean model in the Artificial Analysis frontier index. Upstage is one of four remaining sovereign-AI project teams, took ~KRW 560B in National Growth Fund equity (May 2026), acquired Daum portal operator AXZ from Kakao (June 2026) to build an AI portal, and is preparing a KOSDAQ IPO. Product-specific usage and named-customer scale remain partially disclosed.", "descriptionKo": "Solar는 업스테이지가 개발한 고효율 LLM으로, 작은 모델 크기에서도 높은 성능을 내는 한국어 엔터프라이즈 특화 모델입니다. Solar Mini와 Solar Pro 3 두 가지 크기를 제공하며, 문서 이해, 한국어 우수성, 온프레미스 배포 지원이 주요 강점입니다. Upstage Console, OpenRouter, BizRouter 등을 통해 이용할 수 있으며, Solar Pro 2 사용자는 Pro 3으로 마이그레이션이 지원됩니다. 추론 인프라는 FriendliAI를 사용하는 등 한국 AI 생태계 내 협업 구조도 갖추고 있습니다. 공개된 제품별 유료 고객 수나 도입 규모는 없으나, 글로벌 LLM 벤치마크에서 동급 모델 대비 경쟁력 있는 성적을 유지하고 있습니다.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Upstage's Solar Pro 2 (31B) was the first Korean model recognised as frontier-class by Artificial Analysis — third-party evaluation rather than a vendor claim. Upstage is one of five consortia in Korea's national sovereign foundation-model programme. Product-specific user counts are not disclosed. Korea's Ministry of Science and ICT selected five consortia — LG AI Research, SK Telecom, Naver Cloud, NC AI and Upstage — for the national sovereign foundation-model mandate under a ~$390M programme reviewed every six months. MSIT opened bidding on 14 July 2026 for \"AI for Everyone\", a free national assistant required to run at least 50% on domestic models, with beta targeted for late September 2026.", "releaseDate": "2023-12", "technicalImpact": 80, "competitors": [ "hyperclova-x", "exaone", "kanana", "GPT-5", "claude", "Gemini 2.5", "Grok 4", "qwen" ], "features": [ "Efficient architecture", "Korean excellence", "Document understanding", "On-premise deployment", "Suite modules: Solar Mini, Solar Pro 3", "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization", "Production LLM API", "Improved reasoning", "Stronger instruction following", "Korean and English enterprise workloads" ], "pricing": { "api": "Usage-based", "enterprise": "Custom agreements" }, "link": "https://console.upstage.ai/docs/models/solar", "apiLink": "https://console.upstage.ai", "aiCoreShare": 100, "nativeName": "솔라 LLM", "sourceUrls": [ "https://www.upstage.ai/blog/en/solar-pro-3-0127", "https://console.upstage.ai/docs/models/solar", "https://www.upstage.ai/blog/en/solar-pro-3-0127", "https://www.digitaltoday.co.kr/en/view/53208/upstage-completes-acquisition-of-axz-daum-operator-to-shift-to-ai-portal" ], "scoringRationale": "Reduced from 84 to 80 in the 2026-07-29 fairness pass, applying the adoption rule uniformly: benchmarks plus a sovereign mandate do not reach the tier of adoption evidence (Llama, 500M+ downloads) or global ecosystem roles (GLM, Kimi). Third-party recognition from Artificial Analysis, the sovereign mandate, and Upstage's enterprise/document-AI business are sourced and earn the 80 tier above undisclosed domestic model rows (78). Solar's own prior rationale conceded it lacks the global open-weight distribution of Qwen/DeepSeek/Kimi — that is precisely the 84-tier bar.", "versions": [ { "name": "Solar Pro 2", "released": "2026", "variants": [ "31B parameters" ], "note": "First Korean model recognised as frontier-class by Artificial Analysis." }, { "name": "Solar Pro" }, { "name": "Solar Mini" } ], "versionsUpdated": "2026-07-29" }, { "id": "gemma", "name": "Gemma", "company": "Google", "region": "global", "tagline": "Google's Apache-licensed open model family for local and edge AI", "description": "Gemma is Google's open-weight model family for local, edge and self-hosted AI. Gemma 4 provides Apache 2.0-licensed weights in four deployment sizes. Google's April 2026 release reports over 400M family downloads and 100,000 community variants, not active users. Earlier generations have their own licence terms.", "users": "400M+ family downloads (Google, Apr 2026; not active users)", "userCount": 0, "adoptionSignal": "Google reported more than 400M cumulative family downloads and 100,000 community variants in April 2026. These indicate developer activity, not unique users or Gemma 4-specific adoption.", "releaseDate": "2026-04", "technicalImpact": 84, "competitors": [ "llama", "qwen", "deepseek", "gpt-oss" ], "features": [ "Apache 2.0 open model weights", "E2B and E4B edge models", "26B MoE and 31B dense variants", "Agentic workflows and code generation", "Multimodal input and long context" ], "pricing": { "open_weight": "Free weights", "cloud": "Google Cloud / Vertex AI deployment" }, "link": "https://ai.google.dev/gemma", "apiLink": "https://aistudio.google.com/", "aiCoreShare": 100, "sourceUrls": [ "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "https://ai.google.dev/gemma/docs/releases" ], "scoringRationale": "Reviewed 2026-09-29: retains Impact 84. Current Gemma 4 weights and a broad derivative ecosystem replace the stale 150–200M-download rationale. Downloads remain userCount 0; current market-share leadership is not established.", "versions": [ { "name": "Gemma 4", "released": "2026-04", "variants": [ "E2B (edge)", "E4B (edge)", "26B MoE", "31B dense" ], "note": "Current generation. Apache 2.0 weights, multimodal input, long context, agentic and code workloads." }, { "name": "Gemma 3", "released": "2025-03" } ], "versionsUpdated": "2026-09-29" }, { "id": "kakao-i-connect", "name": "Kakao i", "company": "Kakao", "region": "korea", "tagline": "Kakao's AI platform and business messaging stack", "description": "Kakao i is Kakao's AI platform layer — speech, translation, vision and conversational services — together with the business messaging and call products that let companies reach customers through KakaoTalk, Korea's dominant messenger. Channel access is the defence rather than model quality. Kakao discloses no product-specific customer or volume figure.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "Perplexity", "You.com", "Liner", "Phind" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization", "Suite modules: Kakao i" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://kakaocorp.com", "apiLink": null, "aiCoreShare": 100, "sourceUrls": [], "versions": [ { "name": "Kakao i", "variants": [ "Kakao i" ] } ], "scoringRationale": "78 reflects channel access to KakaoTalk, Korea's dominant messenger, which is a genuine distribution asset for business messaging. Held at the cap because Kakao discloses no product-specific customer count or message volume, and the moat is Kakao's channel rather than the AI itself.", "versionsUpdated": "2026-07-29" }, { "id": "medgemma", "name": "MedGemma", "company": "Google", "region": "global", "tagline": "Healthcare-tuned open model family from Google", "description": "MedGemma is Google's healthcare-tuned open-weight model family, derived from Gemma and released for medical text and imaging research where teams need to self-host rather than send clinical data to an API. It is a research and prototyping asset rather than a cleared clinical product, and Google discloses no deployment figures.", "users": "Not disclosed", "userCount": 0, "releaseDate": "2026-04", "technicalImpact": 78, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://ai.google.dev", "apiLink": "https://ai.google.dev/gemini-api/docs/models", "aiCoreShare": 100, "versions": [ { "name": "MedGemma", "released": "2026-04", "note": "Gemma-derived; released for self-hosted medical research." } ], "versionsUpdated": "2026-07-29" }, { "id": "paligemma", "name": "PaliGemma", "company": "Google", "region": "global", "tagline": "Open vision-language model family from Google", "description": "PaliGemma is Google's open-weight vision-language model family, derived from Gemma and SigLIP, for image captioning, visual question answering and document understanding on self-hosted infrastructure. Google discloses no deployment figures.", "users": "Not disclosed", "userCount": 0, "releaseDate": "2026-04", "technicalImpact": 78, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://ai.google.dev", "apiLink": "https://ai.google.dev/gemini-api/docs/models", "aiCoreShare": 100, "versions": [ { "name": "PaliGemma", "released": "2026-04", "note": "Gemma-derived vision-language weights." } ], "versionsUpdated": "2026-07-29" }, { "id": "baidu-comate", "name": "Baidu Comate", "company": "Baidu", "region": "china", "tagline": "Baidu AI coding assistant", "description": "Baidu Comate is Baidu's AI coding assistant surface. It is a developer workflow product, separate from Baidu's Qianfan model and agent platform.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://comate.baidu.com/", "apiLink": null, "aiCoreShare": 100, "sourceUrls": [ "https://comate.baidu.com/" ], "versions": [ { "name": "Baidu Comate", "note": "Current coding-assistant surface; Baidu does not publish a stable public semantic version." } ], "versionsUpdated": "2026-09-02" }, { "id": "model-studio", "name": "Alibaba Cloud Model Studio", "company": "Alibaba", "region": "china", "tagline": "Qwen model-building and application platform", "description": "Alibaba Cloud Model Studio is the developer platform for building, evaluating, and deploying applications with Qwen and other models. It belongs in the catalog as a platform surface, not as another Qwen model-version row.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.alibabacloud.com/product/modelstudio", "apiLink": "https://bailian.console.aliyun.com/", "aiCoreShare": 100, "sourceUrls": [ "https://www.alibabacloud.com/product/modelstudio", "https://www.alibabacloud.com/help/en/model-studio/models" ], "adoptionSignal": "Current first-party product material names AstraZeneca as a dedicated Model Studio customer; aggregate customer or API usage is not disclosed.", "versions": [ { "name": "Model Studio", "variants": [ "Qwen 3.8", "Wan 3.0", "DeepSeek V4", "GLM 5.2" ], "note": "Current platform model catalog at the September 2026 review." } ], "versionsUpdated": "2026-09-02" }, { "id": "amazon-titan", "name": "Amazon Titan", "company": "AWS", "region": "global", "tagline": "Amazon Bedrock foundation-model family", "description": "Amazon Titan is Amazon's Bedrock model family for image generation and embeddings. Separate embedding or modality rows are variants rather than independent product lines.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "Midjourney", "chatgpt", "Stable Diffusion", "Ideogram", "OpenAI Embeddings", "Voyage AI", "Cohere Embed", "BGE" ], "features": [ "Image generation", "Text embeddings", "Multimodal embeddings", "Amazon Bedrock deployment", "Suite modules: Amazon Titan Text Embeddings V2, Amazon Titan Multimodal Embeddings", "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://aws.amazon.com/bedrock/amazon-models/", "apiLink": "https://docs.aws.amazon.com/bedrock/", "aiCoreShare": 100, "sourceUrls": [ "https://aws.amazon.com/bedrock/titan/" ], "scoringRationale": "Amazon Titan is AWS's legacy first-party model family (Titan Text, Titan Embeddings, Titan Image) predating Nova. Scored at 78 because Titan has been superseded by Amazon Nova as AWS's primary model family and is maintained for backward compatibility rather than as a frontier offering.", "versions": [ { "name": "Amazon Titan", "variants": [ "Image Generator", "Text Embeddings v2", "Multimodal Embeddings" ], "note": "Superseded by Nova for new Bedrock workloads." } ], "versionsUpdated": "2026-07-29" }, { "id": "clova-dubbing", "name": "CLOVA Dubbing", "company": "Naver", "region": "korea", "tagline": "Naver's browser-based AI voice and video dubbing service", "description": "CLOVA Dubbing turns typed scripts into downloadable voice tracks for video and audio production. Naver ended the Android and iOS apps on July 20, 2026 while keeping the PC and mobile web service active.", "users": "100K+ subscribers (2020; current count not disclosed)", "userCount": 100000, "releaseDate": "", "technicalImpact": 74, "competitors": [ "elevenlabs", "Papercup", "Deepdub", "Speechify Dub" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://clovadubbing.naver.com/", "apiLink": null, "aiCoreShare": 100, "scoringRationale": "Held at 74: Naver reported 100,000 subscribers in 2020, but has not disclosed a current count; the web service remains active after the July 2026 mobile-app shutdown.", "adoptionSignal": "Naver reported 100,000 subscribers within four months of the 2020 launch; current notices confirm the web service continues after the mobile apps ended in July 2026.", "sourceUrls": [ "https://clovadubbing.naver.com/", "https://clovadubbing.naver.com/notice", "https://www.navercorp.com/navercorp_/ir/annualReport/2021/NAVER_2020AR_ENG.pdf" ], "versions": [ { "name": "CLOVA Dubbing Web", "note": "PC and mobile web remain active; Android and iOS apps ended on 2026-07-20." } ], "versionsUpdated": "2026-09-02" }, { "id": "coze", "name": "Coze", "company": "ByteDance", "region": "china", "tagline": "ByteDance AI agent and bot-building platform", "description": "Coze is ByteDance's agent and bot-building platform for automating workflows with model calls, tools, memory, and integrations. It is a derived app/platform surface, not another Doubao chat row.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.coze.com/", "apiLink": null, "aiCoreShare": 100, "sourceUrls": [ "https://www.coze.com/" ], "versions": [ { "name": "Coze", "note": "Current agent-building and intelligent-office platform surface." } ], "versionsUpdated": "2026-09-02" }, { "id": "la-plateforme", "name": "Mistral AI Studio", "company": "Mistral AI", "region": "global", "tagline": "Mistral AI production platform for building and operating AI systems", "description": "Mistral AI Studio is the current product surface for what was previously La Plateforme: an enterprise AI production platform for building, evaluating, deploying, and operating applications and agents on Mistral models.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://mistral.ai/products/studio", "apiLink": "https://docs.mistral.ai", "aiCoreShare": 100, "sourceUrls": [ "https://mistral.ai/products/studio" ], "scoringRationale": "La Plateforme (Mistral AI Studio) is Mistral's proprietary API gateway and developer console for accessing Mistral models. Scored at 76 because it is a distribution channel for Mistral's models rather than a standalone AI product — the model-level scores (Mistral Large 81, Mistral Small 78) capture the underlying value. Kept at 76 rather than lower because Mistral's model quality and active developer ecosystem make La Plateforme a real, used API surface, not a placeholder.", "versions": [ { "name": "Mistral AI Studio", "released": "2026", "note": "Current product surface; previously La Plateforme." }, { "name": "La Plateforme" } ], "versionsUpdated": "2026-07-29" }, { "id": "amazon-nova", "name": "Amazon Nova", "company": "AWS", "region": "global", "tagline": "Amazon's multimodal foundation-model family in Bedrock", "description": "Amazon Nova is Amazon's foundation-model family in Bedrock, spanning text, multimodal, image, video, and speech variants. Individual model sizes and modes should resolve to the family-level product row.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4", "AWS Bedrock", "gemini-enterprise-agent-platform", "Azure AI Foundry", "Databricks", "Vapi", "Retell AI" ], "features": [ "Text and multimodal models", "Image, video, and speech generation", "Amazon Bedrock deployment", "Enterprise API access", "Suite modules: Amazon Nova Micro, Amazon Nova Lite, Amazon Nova Premier, Amazon Nova Canvas, Amazon Nova Reel, Amazon Nova Sonic, Amazon Nova 2 Lite, Amazon Nova 2 Sonic", "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization", "Suite modules: Amazon Nova Act", "Agent orchestration", "Tool calling", "Workflow automation", "Enterprise controls" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://aws.amazon.com/ai/generative-ai/nova/", "apiLink": "https://docs.aws.amazon.com/bedrock/", "aiCoreShare": 100, "sourceUrls": [], "scoringRationale": "Amazon Nova is AWS's first-party frontier model family (Nova Micro, Lite, Pro, Premier) available exclusively on Bedrock, designed for cost-efficient enterprise inference. Scored at 78 because Nova is a newer model family without disclosed adoption metrics or benchmark leadership claims relative to Claude, Gemini, or GPT-4o on Bedrock — positioned as cost-efficient alternatives rather than frontier-quality substitutes.", "versions": [ { "name": "Amazon Nova 2", "variants": [ "Nova 2 Lite", "Nova 2 Sonic" ], "note": "Current generation in Bedrock." }, { "name": "Amazon Nova", "variants": [ "Micro", "Lite", "Pro", "Premier", "Canvas", "Reel", "Sonic" ] } ], "versionsUpdated": "2026-07-29" }, { "id": "doubao-seed", "name": "Doubao Seed", "company": "ByteDance", "region": "china", "tagline": "ByteDance's enterprise model family on Volcano Engine Ark", "description": "Doubao Seed is ByteDance's enterprise model family served through Volcano Engine Ark. The family spans reasoning, coding, vision, realtime voice, image, video, speech and embedding variants; modality-specific releases belong in this row rather than separate product pages.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization", "Suite modules: Doubao Seed Vision" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.volcengine.com/product/doubao", "apiLink": "https://www.volcengine.com/product/ark", "aiCoreShare": 100, "sourceUrls": [ "https://www.volcengine.com/product/doubao", "https://www.volcengine.com/product/ark", "https://seed.bytedance.com/en/models?view_from=homepage_tab", "https://seed.bytedance.com/en/blog/seedrealtime-audio-visual-full-duplex-llm-released-toward-omni-modal-natural-interaction" ], "versions": [ { "name": "Doubao Seed 2.1", "variants": [ "2.1 Turbo", "Seed Vision", "Seed RealtimeVoice", "Seed TTS 2.0", "SeedRealtime" ], "note": "Current enterprise model generation and modality variants on Volcano Engine Ark." }, { "name": "Doubao Seed 2.0" }, { "name": "Doubao Seed 1.6" } ], "versionsUpdated": "2026-09-02" }, { "id": "trae-ai-ide", "name": "TRAE", "company": "ByteDance", "region": "china", "tagline": "ByteDance's AI development family spanning TRAE IDE and TRAE Work", "description": "TRAE is ByteDance's AI development product family. TRAE IDE serves software developers and engineering teams, while TRAE Work extends agentic building, research and professional workflows to desktop, web and mobile.", "users": "6M+ registered users (reported for 2025)", "userCount": 6000000, "releaseDate": "", "technicalImpact": 84, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.trae.ai/", "apiLink": null, "aiCoreShare": 100, "sourceUrls": [ "https://www.trae.ai/", "https://www.trae.ai/blog/trae_2025_yearly_review", "https://www.trae.ai/changelog", "https://www.trae.ai/blog/trae_work_0609" ], "adoptionSignal": "TRAE reported 6 million registered users across nearly 200 countries for 2025, plus 365,000 custom agents created or updated.", "scoringRationale": "Raised from 76 to 84 on product-specific adoption evidence: TRAE reported 6 million registered users across nearly 200 countries. It clears the horizontal-tool evidence cap but stays below the 90+ tier because durable workflow control and paid active usage are not disclosed.", "versions": [ { "name": "TRAE", "variants": [ "TRAE IDE 3.5", "TRAE Work", "TRAE App" ], "note": "Current product family after TRAE SOLO evolved into TRAE Work in June 2026." } ], "versionsUpdated": "2026-09-02" }, { "id": "volcano-engine-ark", "name": "Volcano Engine Ark", "company": "ByteDance", "region": "china", "tagline": "ByteDance cloud model and agent platform", "description": "Volcano Engine Ark is ByteDance's cloud platform for model access, model deployment, and agent/application workflows, including Doubao Seed model access.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.volcengine.com/product/ark", "apiLink": "https://www.volcengine.com/product/ark", "aiCoreShare": 100, "sourceUrls": [ "https://www.volcengine.com/product/ark" ], "versions": [ { "name": "Volcano Engine Ark", "variants": [ "Doubao Seed 2.1", "Doubao Seed Evolving", "Ark Agent Plan" ], "note": "Current model, inference and agent platform surface." } ], "versionsUpdated": "2026-09-02" }, { "id": "le-chat", "name": "Vibe", "company": "Mistral AI", "region": "global", "tagline": "Mistral AI assistant and long-horizon work agent, formerly Le Chat", "description": "Vibe is Mistral AI's AI assistant and agent product for work and coding. Mistral describes Vibe as the successor to Le Chat, with connected tools, deep research, coding workflows, and enterprise deployment options.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "AI chat and work assistant", "Deep research mode", "Coding and agentic workflows", "Connected tools and enterprise deployment" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://mistral.ai/products/vibe/", "apiLink": "https://docs.mistral.ai", "aiCoreShare": 100, "sourceUrls": [ "https://mistral.ai/products/vibe/" ], "scoringRationale": "Vibe (formerly Le Chat) is Mistral AI's consumer and work AI assistant, rebranded from Le Chat to emphasize agentic coding and deep-research capabilities. Competes directly with ChatGPT, Claude, and Gemini in the AI assistant category. Scored at 76 because active user counts are not disclosed and Vibe is early-stage as a rebranded product — it has not yet demonstrated the retention and adoption scale of the top-tier consumer AI assistants it competes with. Score re-evaluates upward if Mistral discloses MAU, subscription revenue, or enterprise seat counts for Vibe specifically.", "versions": [ { "name": "Vibe", "released": "2026", "variants": [ "Connected tools", "Deep research", "Coding workflows" ], "note": "Mistral describes Vibe as the successor to Le Chat." }, { "name": "Le Chat" } ], "versionsUpdated": "2026-07-29" }, { "id": "segment-anything-model-2", "name": "Segment Anything (SAM)", "company": "Meta", "region": "global", "tagline": "Open and commercial AI model or AI product from Meta", "description": "Segment Anything Model 2 is Meta's open/commercial model and assistant surface.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 79, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://llama.meta.com", "apiLink": "https://ai.meta.com", "aiCoreShare": 100, "scoringRationale": "Meta's SAM 2 is the second-generation foundation model for promptable image and video segmentation, open-weight and widely used in computer vision research and production pipelines. Scored at 79 reflecting strong research adoption and open-weight usage; kept below 80-84 products that have commercial deployment evidence.", "sourceUrls": [ "https://ai.meta.com/sam2/" ], "versions": [ { "name": "SAM 2", "released": "2024-07", "note": "Adds video segmentation to the image model." }, { "name": "SAM", "released": "2023-04" } ], "versionsUpdated": "2026-07-29", "contentReview": "needs-review" }, { "id": "tencent-cloud-ti-platform", "name": "Tencent Cloud TI-ONE", "company": "Tencent", "region": "china", "tagline": "Tencent Cloud's end-to-end machine-learning platform", "description": "Tencent Cloud TI-ONE is Tencent's managed machine-learning platform for data access, model training, model management and inference services. It is the enterprise MLOps counterpart to Alibaba Cloud Model Studio and Baidu Qianfan; Tencent does not disclose a platform-specific customer count.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 80, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www-sg.tencentcloud.com/document/product/1141/73398", "apiLink": "https://www-sg.tencentcloud.com/document/product/1141/73398", "aiCoreShare": 100, "scoringRationale": "Tencent Cloud TI-ONE is a maintained end-to-end MLOps surface, but stays at 80 because Tencent does not disclose product-specific customer or usage scale.", "sourceUrls": [ "https://www-sg.tencentcloud.com/document/product/1141/73398" ], "versions": [ { "name": "TI-ONE", "note": "Current Tencent Cloud machine-learning platform and API surface." } ], "versionsUpdated": "2026-09-02" }, { "id": "konan-search", "name": "Konan Technology", "company": "Konan Technology", "region": "korea", "tagline": "Korean-language enterprise search, LLM and analytics for closed networks", "description": "Konan Technology supplies Korean government agencies and large corporates with enterprise search, conversational AI, media monitoring and analytics built around Korean morphological analysis, deployed on-premises inside closed networks where public assistants cannot operate. That deployment constraint is the moat rather than model quality. Konan is KOSDAQ-listed with long-standing public-sector contracts; per-product customer counts are not disclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "Glean", "Cohere RAG", "Vectara", "Mendable" ], "features": [ "Korean-language enterprise search", "On-premises closed-network deployment", "Conversational AI and chatbot", "Media monitoring and analytics", "Public-sector procurement track record", "Suite modules: Konan Chatbot, Konan BI, Konan LLM, Konan Watcher", "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://konantech.com", "apiLink": null, "aiCoreShare": 100, "sourceUrls": [], "versions": [ { "name": "Konan Technology", "variants": [ "Konan Chatbot", "Konan BI", "Konan LLM", "Konan Watcher" ] } ], "adoptionSignal": "Konan LLM ENT-11 current enterprise flagship; open Konan-LLM-OND on Hugging Face; applied to but not selected for the sovereign-AI project", "versionsUpdated": "2026-07-29" }, { "id": "baidu-ai-cloud-qianfan", "name": "Baidu Qianfan", "company": "Baidu", "region": "china", "tagline": "Agent-centered enterprise model platform from Baidu AI Cloud", "description": "Baidu Qianfan is an enterprise large-model platform centered on agents, multi-agent orchestration, workflow agents, knowledge-base RAG, observability, model services, MCP services, and private deployment options.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 82, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://cloud.baidu.com/product-s/qianfan_home", "apiLink": "https://cloud.baidu.com/product-s/qianfan_home", "aiCoreShare": 100, "nativeName": "千帆大模型平台", "sourceUrls": [ "https://cloud.baidu.com/product-s/qianfan_home", "https://intl.cloud.baidu.com/en/product/qianfan.html" ], "adoptionSignal": "Baidu names Qianfan deployments serving 200,000 households in a local-government assistant and supporting a national trade-union system whose potential reach spans hundreds of millions of members; actual platform users and API volume are not disclosed.", "scoringRationale": "Raised from 77 to 82 on third-party-facing named deployment evidence published in Baidu's current product case studies. Qianfan supports a 200,000-household government assistant and a nationwide trade-union service, but user activity and API volume remain undisclosed, so it stays below globally standardized cloud platforms.", "versions": [ { "name": "Baidu Qianfan", "variants": [ "Agent engine", "ModelBuilder", "AppBuilder", "DataBuilder", "MCP services" ] } ], "versionsUpdated": "2026-09-02" }, { "id": "apollo-go-ai", "name": "Apollo Intelligent Driving", "company": "Baidu", "region": "china", "tagline": "Baidu autonomous-driving and intelligent-vehicle platform", "description": "Apollo Intelligent Driving is Baidu's autonomous-driving and intelligent-vehicle product family. The row is kept under physical AI because the model capability is coupled to vehicles, sensors, and operating environments.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.apollo.auto/apollo-self-driving", "apiLink": null, "aiCoreShare": 100, "nativeName": "百度 Apollo", "sourceUrls": [ "https://www.apollo.auto/apollo-self-driving" ], "versions": [ { "name": "Apollo Intelligent Driving", "note": "Current autonomous-driving and intelligent-vehicle product family." } ], "versionsUpdated": "2026-09-02" }, { "id": "clova-studio", "name": "CLOVA Studio", "company": "Naver", "region": "korea", "tagline": "Naver Cloud's AI platform for HyperCLOVA X, OCR and speech", "description": "CLOVA Studio is Naver Cloud's developer platform for building on HyperCLOVA X, alongside the CLOVA OCR and speech APIs that share the same account, billing and deployment path. Korean enterprises use it where Korean-language quality and domestic data residency matter more than frontier capability. Naver discloses no developer or customer count for the platform.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 76, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "HyperCLOVA X tuning and deployment", "OCR for Korean documents and forms", "Speech recognition and synthesis", "Korean data residency", "Naver Cloud Platform integration", "Suite modules: CLOVA OCR, CLOVA Speech", "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://clova.ai", "apiLink": "https://www.ncloud.com/product/aiService/clovaStudio", "aiCoreShare": 100, "nativeName": "클로바 스튜디오", "sourceUrls": [], "versions": [ { "name": "CLOVA Studio", "variants": [ "CLOVA OCR", "CLOVA Speech" ] } ], "scoringRationale": "76 reflects Naver Cloud's developer platform for HyperCLOVA X with Korean data residency, which is the buying reason for Korean regulated enterprises. Held at the default tier because Naver discloses no developer or customer count for the platform, and the model row (HyperCLOVA X, 84) carries the family's evidence.", "versionsUpdated": "2026-07-29" }, { "id": "abab6-5", "name": "MiniMax API", "company": "MiniMax", "region": "china", "tagline": "MiniMax model API and token plan", "description": "MiniMax API is the developer access surface for current MiniMax models, including M3, M2.7, Hailuo, Speech, and Music capabilities. This replaces the stale abab6.5 model-version row.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 80, "competitors": [ "together-inference", "openrouter", "aws-bedrock", "groqcloud", "SiliconFlow MaaS" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://platform.minimax.io/", "apiLink": "https://platform.minimax.io/", "aiCoreShare": 100, "scoringRationale": "Held at 80. MiniMax's public reporting aggregates more than two million enterprises and developers across its open platform and product portfolio, so the bundled figure cannot be assigned to this API row; the flagship MiniMax M family carries the stronger model-tier score.", "sourceUrls": [ "https://www.minimax.io/", "https://platform.minimax.io/", "https://platform.minimax.io/docs/release-notes/models", "https://ir.minimax.io/" ], "adoptionSignal": "MiniMax reports more than two million enterprises and developers across its open platform and products; this is bundled portfolio exposure, not API-specific adoption.", "versions": [ { "name": "MiniMax API", "variants": [ "MiniMax M3", "MiniMax M2.7", "Speech 2.8", "Hailuo 2.3", "Music 3.0" ], "note": "Current multimodal API catalog at the September 2026 review." } ], "versionsUpdated": "2026-09-02" }, { "id": "mistral", "name": "Mistral", "company": "Mistral AI", "region": "global", "tagline": "Mistral AI's general-purpose language model family", "description": "Mistral is Mistral AI's general-purpose language model family, offered in Small and Large size tiers through La Plateforme, cloud marketplaces and open weights. Size tiers are how one family is priced, not separate products, so they share a row — the same treatment Claude, Gemini and Llama get. Mistral AI Studio (the developer platform) and Vibe (the assistant) stay separate rows because they are different purchases for different buyers.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 81, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "General-purpose language model family", "Small and Large size tiers", "Open-weight and commercial licensing", "La Plateforme API and cloud marketplace distribution", "European data-residency positioning", "Suite modules: Mistral Small", "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization", "Suite modules: Mistral Small 3.1" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://mistral.ai/models", "apiLink": "https://docs.mistral.ai", "aiCoreShare": 100, "sourceUrls": [ "https://mistral.ai/technology/" ], "scoringRationale": "Mistral Large is Mistral AI's flagship enterprise closed model (Mistral Large 2 and successors), available via AWS Bedrock, Azure AI Marketplace, and Mistral's own La Plateforme API, with a Microsoft partnership for co-development and distribution. Increased from 78 to 81 to correct underscoring relative to comparable regional foundation models: StepFun Step 3 (80) has no major international cloud marketplace listings and China-primary distribution; Solar LLM (82) is on its own Korean-focused API only. Mistral Large's Bedrock + Azure presence gives it materially broader international enterprise reach than either, plus EU sovereign AI relevance (strong regulatory alignment with European data residency). Stays at 81 rather than 82+ because Mistral-specific user counts, ARR, and enterprise contract volumes are not publicly disclosed, and Solar holds a 1-point premium for Korean-language specialization depth within its target segment.", "versions": [ { "name": "Mistral", "variants": [ "Mistral Small", "Mistral Large" ], "note": "Size tiers of one general-purpose family. Mistral AI Studio and Vibe are catalogued separately." } ], "versionsUpdated": "2026-07-29" }, { "id": "minimax-m1", "name": "MiniMax M", "company": "MiniMax", "region": "china", "tagline": "MiniMax's M-series reasoning and agentic model family", "description": "M is MiniMax's flagship language-model series for coding, reasoning, native multimodal understanding, and long-horizon agent work. M3 supports a 1M-token context and is available through MiniMax's API and coding integrations; numbered releases remain versions of this one series.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 87, "competitors": [ "deepseek", "kimi", "qwen", "claude", "glm-4", "hunyuan" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization", "Suite modules: MiniMax M2.7" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.minimax.io/", "apiLink": "https://platform.minimax.io/", "aiCoreShare": 100, "scoringRationale": "MiniMax M3 is the flagship coding and agentic model surface with long-context positioning and API access. It is scored as MiniMax's strongest current model product while remaining below frontier leaders with stronger disclosed adoption.", "sourceUrls": [ "https://www.minimax.io/", "https://ir.minimax.io/news-events/new-releases" ], "versions": [ { "name": "MiniMax M3", "released": "2026-06", "variants": [ "Coding and agentic workloads", "1M-token context" ] }, { "name": "MiniMax M2.7" } ], "versionsUpdated": "2026-09-02" }, { "id": "minimax-speech", "name": "MiniMax Speech", "company": "MiniMax", "region": "china", "tagline": "MiniMax speech model and audio product line", "description": "MiniMax Speech 2.8 is the current speech model and audio product line in MiniMax's model stack.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 80, "competitors": [ "elevenlabs", "OpenAI TTS", "Google Cloud Text-to-Speech" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.minimax.io/", "apiLink": "https://platform.minimax.io/", "aiCoreShare": 100, "scoringRationale": "MiniMax Speech 2.8 is a real modality product in the MiniMax stack, but no product-specific usage, customer, or revenue scale is disclosed.", "sourceUrls": [ "https://www.minimax.io/" ], "versions": [ { "name": "MiniMax Speech 2.8" } ], "versionsUpdated": "2026-07-29" }, { "id": "kimi", "name": "Kimi", "company": "Moonshot AI", "region": "china", "tagline": "Moonshot AI's assistant and open-weight K-series frontier models", "description": "Kimi is Moonshot AI's assistant and model family, spanning chat, coding, agent workflows, deep research, slides, docs and the Kimi Platform API. Kimi K3, released July 2026, is a 2.8-trillion-parameter multimodal reasoning model with a 1M-token context and native vision — the largest open-weight model released to date, with full weights published on Hugging Face on 27 July 2026. It ranked first in Arena's Frontend Code evaluation at 1,679 points, ahead of Claude Fable 5, in blind developer testing; Moonshot's own benchmarks put it ahead of Claude Opus 4.8 and GPT-5.5 while behind Fable 5 and GPT-5.6 Sol overall.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 90, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Frontier open-weight reasoning models (K-series)", "1M-token context with native vision", "Coding, agent-swarm and deep-research workflows", "Slides, docs, sheets and website generation", "Kimi Platform API with cache-aware pricing" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.kimi.com/", "apiLink": "https://platform.kimi.ai/", "aiCoreShare": 100, "nativeName": "Kimi 智能助手", "sourceUrls": [ "https://www.kimi.com/", "https://platform.kimi.ai/", "https://www.kimi.com/en/blog/kimi-k3", "https://www.kimi.com/news/kimi-k3-open-source" ], "adoptionSignal": "Moonshot published Kimi K3's full 2.8T-parameter weights on Hugging Face on 27 July 2026 with an open licence, and K3 leads Arena's Frontend Code evaluation. Moonshot still discloses no product-specific active-user count for the Kimi assistant.", "scoringRationale": "2026-07-29: raised 88 -> 90, parity with Qwen and DeepSeek. The 90 tier is defined by frontier-model role plus global open-weight ecosystem, and Kimi now fully qualifies on both: K3 (July 2026, weights public on Hugging Face) is the largest open-weight model released to date, ranks #1 in Arena Frontend Code ahead of Claude Fable 5, and sits at near-parity with the closed frontier overall by Moonshot's own benchmarks; the K2 line had already built genuine global developer-ecosystem reach. The prior 2-point gap below Qwen/DeepSeek rested on ecosystem maturity, which the K3 release closes as a matter of role. Not above 90: overall capability remains behind Fable 5 and GPT-5.6 Sol by Moonshot's own numbers, and no active-user figure is disclosed.", "versions": [ { "name": "Kimi K3", "released": "2026-07", "variants": [ "2.8T parameters", "1M context", "Native vision", "Open weights (Hugging Face, 27 Jul 2026)" ], "note": "Largest open-weight model released to date. #1 in Arena Frontend Code (1,679) ahead of Claude Fable 5." }, { "name": "Kimi K2.6" }, { "name": "Kimi K2 / K2 Thinking" } ], "versionsUpdated": "2026-09-02" }, { "id": "tencent-cloud-ai-digital-human", "name": "Tencent Cloud AI Digital Human", "company": "Tencent", "region": "china", "tagline": "Tencent digital-human and avatar product surface", "description": "Tencent Cloud AI Digital Human is the digital-human product surface in Tencent's enterprise AI stack. It is cataloged as a vertical media/enterprise application rather than a Hunyuan model version.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 73, "competitors": [ "heygen", "synthesia", "d-id-agents", "deepbrain-ai-kiosk", "colossyan-ai-avatars" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://intl.cloud.tencent.com/products/ivh", "apiLink": null, "aiCoreShare": 100, "sourceUrls": [ "https://intl.cloud.tencent.com/products/ivh" ], "versions": [ { "name": "Tencent Cloud AI Digital Human", "variants": [ "2D live-action", "3D realistic", "3D stylized", "Interactive avatars" ] } ], "versionsUpdated": "2026-09-02" }, { "id": "minimax-music", "name": "MiniMax Music", "company": "MiniMax", "region": "china", "tagline": "MiniMax music-generation product line", "description": "MiniMax Music is MiniMax's music-generation family. Music 3.0 is the current open-weight generation for complete-song composition, arrangement, vocals, and production from a concept and optional lyrics.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "suno", "udio-remix", "elevenlabs", "Soundraw" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.minimax.io/audio/music", "apiLink": "https://platform.minimax.io/", "aiCoreShare": 97, "scoringRationale": "MiniMax Music 3.0 is a maintained, open-weight production music model, but no product-specific adoption or commercial usage is disclosed, so it remains below music generators with measured reach.", "sourceUrls": [ "https://www.minimax.io/audio/music", "https://ir.minimax.io/news-events/new-releases" ], "versions": [ { "name": "MiniMax Music 3.0", "released": "2026-08", "note": "Open-weight current generation for end-to-end song creation." }, { "name": "MiniMax Music 2.6", "released": "2026-04" } ], "versionsUpdated": "2026-09-02" }, { "id": "step", "name": "Step", "company": "StepFun", "region": "china", "tagline": "StepFun's Step foundation-model family", "description": "Step is StepFun's foundation-model family, released as numbered generations with reasoning and low-latency Flash variants. Individual generations are versions of one family rather than separate products. StepFun does not disclose usage figures; the StepFun API remains a separate row as the developer platform.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 80, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model access", "Fine-tuning or deployment", "API integration", "Language or domain specialization", "Suite modules: Step-3.5-Flash" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.stepfun.com/", "apiLink": null, "aiCoreShare": 100, "sourceUrls": [ "https://www.stepfun.com/", "https://platform.stepfun.com/docs/zh/welcome" ], "scoringRationale": "Current StepFun multimodal foundation-model surface with source-backed public model coverage. It belongs above generic long-tail rows, but stays below MiniMax/Kimi because adoption, API usage, and customer scale are not disclosed.", "versions": [ { "name": "Step 3.7 Flash", "variants": [ "Step 3.5 Flash", "Step-3.5-Flash" ], "note": "Current efficient multimodal model for agent workflows; StepFun API is catalogued separately." } ], "versionsUpdated": "2026-09-02" }, { "id": "hailuo-ai", "name": "Hailuo", "company": "MiniMax", "region": "china", "tagline": "MiniMax AI video-generation product", "description": "Hailuo 2.3 is MiniMax's current video-generation product family, including Hailuo 2.3 and Hailuo 2.3 Fast. It is the canonical MiniMax video surface in the catalog.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 79, "competitors": [ "kling", "veo", "runway", "pika" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://hailuoai.video/", "apiLink": "https://platform.minimax.io/", "aiCoreShare": 100, "scoringRationale": "2026-07-29 global/China top-tier stress test: reduced 84 -> 79. The row's own rationale says it belongs below Kling/Runway-style rows with stronger recorded traction — yet it sat above Runway (80, 5M+ disclosed). With no disclosed usage it joins the undisclosed video-generation tier alongside Pika and Ideogram (79).", "sourceUrls": [ "https://www.minimax.io/", "https://hailuoai.video/" ], "versions": [ { "name": "Hailuo 2.3", "note": "Current MiniMax video-generation release." } ], "versionsUpdated": "2026-07-29" }, { "id": "maas-platform", "name": "Z.ai Open Platform", "company": "Zhipu AI", "region": "china", "tagline": "Z.ai API and model platform", "description": "Z.ai Open Platform is the API and model platform for GLM models and agent workflows. It replaces the older generic MaaS Platform naming in the catalog.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 80, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://open.bigmodel.cn", "apiLink": "https://open.bigmodel.cn", "aiCoreShare": 100, "sourceUrls": [ "https://open.bigmodel.cn", "https://z.ai/" ], "scoringRationale": "Z.ai Open Platform is the developer-facing access layer for GLM models and agents. It has platform relevance, but remains below top-tier frontier platforms until public developer usage, revenue, or customer scale is recorded.", "versions": [ { "name": "Z.ai Open Platform", "note": "API and model platform for GLM models and agent workflows." } ], "versionsUpdated": "2026-07-29" }, { "id": "cohere-embed", "name": "Cohere Platform", "company": "Cohere", "region": "global", "tagline": "Embedding, reranking and retrieval tooling for enterprise RAG", "description": "Cohere's retrieval stack — multilingual embeddings, a reranker that reorders candidate passages, and the Compass and Toolkit developer surfaces — sold as the enterprise RAG layer alongside its Command models. Private and on-premises deployment is the buying reason.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 70, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization", "Suite modules: Cohere Rerank, Cohere Compass, Cohere Toolkit" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://cohere.com", "apiLink": "https://docs.cohere.com", "aiCoreShare": 100, "sourceUrls": [], "versions": [ { "name": "Cohere Platform", "variants": [ "Cohere Rerank", "Cohere Compass", "Cohere Toolkit" ] } ], "versionsUpdated": "2026-07-29" }, { "id": "stepfun-api", "name": "StepFun API", "company": "StepFun", "region": "china", "tagline": "Developer access for StepFun models", "description": "StepFun API is the developer access surface for StepFun models and multimodal capabilities. Public customer counts are not disclosed in the catalog.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model access", "Fine-tuning or deployment", "API integration", "Language or domain specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://platform.stepfun.com/docs/zh/welcome", "apiLink": "https://platform.stepfun.com/docs/zh/api-reference/chat/messages-create", "aiCoreShare": 100, "sourceUrls": [ "https://www.stepfun.com/", "https://platform.stepfun.com/docs/zh/welcome", "https://platform.stepfun.com/docs/zh/step-plan/integrations/reasoning-api" ], "scoringRationale": "Developer access for StepFun models should score as a real model-platform surface, but below the flagship model row because public API adoption and enterprise usage are not disclosed.", "versions": [ { "name": "StepFun API", "variants": [ "Step 3.7 Flash", "Step 3.5 Flash", "Step Router V1", "StepAudio 2.5" ], "note": "Current API and Step Plan model catalog." } ], "versionsUpdated": "2026-09-02" }, { "id": "glm-4", "name": "GLM", "company": "Zhipu AI", "region": "china", "tagline": "Z.ai (Zhipu) GLM model family for agentic engineering", "description": "GLM is Z.ai/Zhipu's foundation-model family for engineering agents, long-horizon work, multimodal understanding, and professional workflows. GLM-5.3-Flash is the current generation, using a hybrid sparse/linear-attention architecture and native multimodal training.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 84, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization", "Suite modules: GLM-4.5, GLM-Z1, ChatGLM" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://docs.z.ai/guides/vlm/glm-5.3-flash", "apiLink": "https://docs.z.ai/", "aiCoreShare": 100, "nativeName": "智谱 GLM", "sourceUrls": [ "https://z.ai/", "https://autoclaw.z.ai/blog/model/glm-5.3-flash/", "https://z.ai/blog/glm-5.2", "https://chat.z.ai/" ], "scoringRationale": "Current Z.ai/Zhipu model surface for agentic engineering and long-horizon tasks. It should score with serious Chinese foundation-model products, but below DeepSeek and top global frontier products because disclosed usage and customer scale are weaker in the catalog.", "versions": [ { "name": "GLM-5.3-Flash", "released": "2026-08", "variants": [ "320B total / 18B active parameters", "Native multimodal", "Hybrid sparse and linear attention" ], "note": "Current efficient generation for agentic, visual and professional workflows." }, { "name": "GLM-5.2", "released": "2026-06" }, { "name": "GLM-4.6" } ], "versionsUpdated": "2026-09-02" }, { "id": "varco-llm", "name": "VARCO", "company": "NCSOFT", "region": "korea", "tagline": "NC AI's game/content foundation-AI platform with a physical-AI pivot", "description": "VARCO is the foundation-AI platform of NC AI (NC Corporation's AI subsidiary), spanning VARCO LLM, VARCO Game AI, and VARCO 3D (version 2.0 released June 24, 2026, +40.8% on Uni3D). After being dropped from Korea's sovereign LLM project in the January 2026 cut, NC AI pivoted to lead the 53-member K-Physical AI Alliance for a government-backed physical-AI world foundation model, building on VARCO 3D and large-scale game simulation. External customer adoption beyond NC's own game production remains undisclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://ncsoft.github.io/ncresearch/varco/", "apiLink": null, "aiCoreShare": 100, "nativeName": "바르코 LLM", "sourceUrls": [ "https://ncsoft.github.io/ncresearch/varco/", "https://en.sedaily.com/news/2026/06/24/nc-ai-unveils-varco-3d-20-for-distortion-free-3d-asset", "https://www.koreaherald.com/article/10674659" ], "adoptionSignal": "NC AI's VARCO is one of five model families selected for Korea's national sovereign foundation-model programme alongside HyperCLOVA X, EXAONE, A.X and Solar, and is required to serve at least half of inference on the 'AI for Everyone' national assistant. NC AI discloses no model-specific adoption figure.", "versions": [ { "name": "VARCO", "released": "2026", "note": "One of five model families in Korea's national sovereign foundation-model programme." } ], "scoringRationale": "Raised from 69 to 78 on cross-vendor consistency. VARCO holds the same national sovereign foundation-model mandate as HyperCLOVA X, EXAONE, A.X and Solar, and MSIT requires selected platforms to route at least half of inference through those five families — scoring it nine points below the lowest of its co-mandated peers reflected thin cataloguing rather than weaker standing. Held at 78 rather than 84 because VARCO has no third-party frontier benchmark recognition and NC AI discloses no adoption, unlike the three rows at 84.", "versionsUpdated": "2026-07-29" }, { "id": "jamba", "name": "Jamba", "company": "AI21 Labs", "region": "global", "tagline": "AI21 hybrid SSM-Transformer model family", "description": "Jamba is AI21's hybrid Mamba-Transformer model family for private deployment and long-context text workflows. Documentation lists Jamba 2 Mini and 3B alongside Jamba Large 1.7. Older 1.5 API snapshots are deprecated; their research page is not the current product entry point.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 67, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Hybrid Mamba-Transformer architecture", "Private model deployment", "Long-context text processing", "Jamba 2 Mini and 3B; Jamba Large 1.7" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://docs.ai21.com/docs/jamba-foundation-models", "apiLink": "https://docs.ai21.com/docs/jamba-foundation-models", "aiCoreShare": 100, "sourceUrls": [ "https://docs.ai21.com/docs/jamba-foundation-models" ], "versions": [ { "name": "Jamba 2", "released": "2026-01", "variants": [ "Mini", "3B" ], "note": "Current small variants; Large remains version 1.7." }, { "name": "Jamba Large 1.7", "released": "2025-07" } ], "versionsUpdated": "2026-09-29", "scoringRationale": "Reviewed 2026-09-29: retains 67. Current access verified; no new adoption proof justifies an increase." }, { "id": "command-r", "name": "Cohere Command", "company": "Cohere", "region": "global", "tagline": "Cohere's enterprise model family for RAG and tool use", "description": "Command is Cohere's enterprise language-model family, built for retrieval-augmented generation, tool use and multilingual business workloads, offered in R, R+ and A tiers on one API. Cohere positions it for private and on-premises deployment where data cannot leave the customer environment.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 67, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization", "Suite modules: Command R+", "Suite modules: Command A" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://docs.cohere.com/docs/command-r", "apiLink": "https://docs.cohere.com", "aiCoreShare": 100, "sourceUrls": [ "https://docs.cohere.com/docs/command-r", "https://docs.cohere.com/docs/command-r-plus", "https://docs.cohere.com/docs/command-a" ], "versions": [ { "name": "Command R", "variants": [ "Command R", "Command R+" ], "note": "Enterprise RAG and tool-use model tiers." } ], "versionsUpdated": "2026-07-29" }, { "id": "minicpm", "name": "MiniCPM", "company": "ModelBest", "region": "china", "tagline": "ModelBest's efficient multimodal model family for edge deployment", "description": "MiniCPM is ModelBest's efficient edge-model family for phones, AI PCs, vehicles, robots and wearables. Language, vision and omni releases are variants of one deployable family rather than separate products or a separate platform row.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 78, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model access", "Fine-tuning or deployment", "API integration", "Language or domain specialization", "Suite modules: MiniCPM-V, MiniCPM-o, ModelBest Edge AI Platform" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://github.com/OpenBMB/MiniCPM", "apiLink": null, "aiCoreShare": 100, "sourceUrls": [ "https://modelbest.cn/en/", "https://github.com/OpenBMB/MiniCPM", "https://www.modelbest.cn/", "https://github.com/OpenBMB/MiniCPM-V", "https://github.com/OpenBMB/MiniCPM-o" ], "scoringRationale": "Efficient on-device language model with open-model evidence. It is useful strategically, but lower than MiniCPM-V and larger model platforms because adoption and deployment scale are not recorded.", "versions": [ { "name": "MiniCPM 4.1", "variants": [ "MiniCPM-V 4.5", "MiniCPM-o 2.6" ], "note": "Current language family with vision and omni variants for edge deployment." } ], "versionsUpdated": "2026-09-02" }, { "id": "exaone", "name": "EXAONE", "company": "LG AI Research", "region": "korea", "tagline": "Korean foundation-model program with internal LG deployment and open-weight releases", "description": "EXAONE is LG AI Research's foundation-model family. K-EXAONE (236B mixture-of-experts, 23B active) is the sovereign-project frontier model — the only non-US/China model in the global open-weight top 10 at release — alongside EXAONE 4.5 (33B vision-language, April 2026). Portal Zum fully adopted K-EXAONE for AI search on June 29, 2026, the first commercial deployment of a Korean sovereign-project model, and Elice joined the LG consortium to operationalize deployment. Broad external adoption beyond these named deployments is not yet disclosed; LG Group scale should not be counted as model adoption.", "descriptionKo": "EXAONE는 LG AI 리서치가 개발한 파운데이션 모델 패밀리입니다. EXAONE 3.5, EXAONE 4.5, EXAONE Deep 등 다양한 버전이 있으며, 한국어-영어 이중 언어 모델과 오픈웨이트 릴리스를 제공합니다. ChatEXAONE는 LG 그룹 임직원을 대상으로 내부 배포되어 실제 사용 사례가 있으나, LG 그룹 규모를 EXAONE 사용자 수로 계산하지는 않습니다. 기업·연구기관 협업도 진행 중이지만 외부 고객 수나 유료 도입 규모는 공개되지 않았습니다. 기술적으로 경쟁력 있는 오픈웨이트 모델이지만, 배포 증거 면에서는 Llama·Qwen 같은 글로벌 오픈웨이트 모델에 비해 뒤처져 있습니다.", "users": "Not disclosed", "userCount": 0, "releaseDate": "2023-07", "technicalImpact": 80, "competitors": [ "hyperclova-x", "solar-llm", "kanana", "Llama 4", "Mistral", "qwen", "DeepSeek V3", "GPT-5", "claude", "Gemini 2.5" ], "features": [ "Korean-English bilingual models", "Open-weight model releases", "Enterprise and research collaborations", "Vision-language and reasoning variants" ], "pricing": { "enterprise": "Contact LG AI Research" }, "link": "https://www.lgresearch.ai/exaone", "apiLink": null, "aiCoreShare": 100, "adoptionSignal": "LG AI Research released EXAONE 4.5 in April 2026, a multimodal vision-language generation. EXAONE is one of five model families selected for Korea's national sovereign foundation-model programme. LG discloses no product-specific user or deployment count. Korea's Ministry of Science and ICT selected five consortia — LG AI Research, SK Telecom, Naver Cloud, NC AI and Upstage — for the national sovereign foundation-model mandate under a ~$390M programme reviewed every six months. MSIT opened bidding on 14 July 2026 for \"AI for Everyone\", a free national assistant required to run at least 50% on domestic models, with beta targeted for late September 2026.", "developmentSpend": "Not disclosed; multi-year corporate AI research program launched in 2020.", "capitalEfficiency": "Unproven externally: meaningful internal deployment, but no disclosed external user count or revenue.", "scoringRationale": "Reduced from 84 to 80 in the 2026-07-29 fairness pass alongside HyperCLOVA X and Solar — same evidence class, same rule: Korea-leading standing, K-EXAONE sovereign mandate and the Zum deployment are sourced and earn the 80 tier, but with no disclosed adoption and no global developer ecosystem the row cannot sit at the 84 tier of Llama-scale downloads or GLM/Kimi-scale ecosystem reach.", "sourceUrls": [ "https://www.lgcorp.com/media/release/28482", "https://huggingface.co/LGAI-EXAONE", "https://github.com/LG-AI-EXAONE/EXAONE-3.5", "https://arxiv.org/abs/2604.08644", "https://en.prnasia.com/releases/global/lg-unveils-new-ai-model-exaone-deep--482078.shtml", "https://www.lgresearch.ai/exaone", "https://en.sedaily.com/technology/2026/06/29/portal-zum-adopts-k-exaone-in-ai-search-first-commercial", "https://www.prnewswire.com/news-releases/lg-rolls-outs-k-exaone-south-korea-joins-the-global-frontier-ai-race-with-world-class-ai-model-302658264.html" ], "nativeName": "엑사원", "versions": [ { "name": "EXAONE 4.5", "released": "2026-04", "variants": [ "Vision-language model", "Unified vision encoder + LLM" ], "note": "Current generation. Multimodal reasoning across text and images." }, { "name": "EXAONE 4.0", "released": "2025", "note": "Selected for Korea's national sovereign-AI programme." }, { "name": "EXAONE 3.5" } ], "versionsUpdated": "2026-07-29" }, { "id": "cogview", "name": "CogView", "company": "Zhipu AI", "region": "china", "tagline": "Z.ai vision and image-generation model surface", "description": "Z.ai's vision and image-generation work has moved beyond the older CogView label into newer GLM-4.6V and GLM-5-era multimodal releases. The route remains for continuity but now points at the current Z.ai surface.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 74, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://huggingface.co/zai-org/GLM-4.6V", "apiLink": "https://docs.z.ai/", "aiCoreShare": 100, "sourceUrls": [ "https://z.ai/", "https://huggingface.co/zai-org/GLM-4.6V", "https://chat.z.ai/" ], "scoringRationale": "Z.ai current vision/image model surface is relevant as part of GLM multimodal coverage, but it is not scored like the flagship GLM row because separate adoption and deployment evidence are limited.", "versions": [ { "name": "GLM-4.6V / CogView", "note": "Z.ai vision and image-generation surface." } ], "versionsUpdated": "2026-07-29" }, { "id": "baichuan-2", "name": "Baichuan Model Platform", "company": "Baichuan AI", "region": "china", "tagline": "Baichuan's model API and enterprise agent platform", "description": "Baichuan Model Platform is Baichuan AI's current developer and enterprise surface for the Baichuan model family and agent workflows. The API, model-version and generic assistant rows represented one buyer-facing platform and now share this canonical page.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 69, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model access", "Fine-tuning or deployment", "API integration", "Language or domain specialization", "Suite modules: Baichuan API, Baichuan Assistant" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://platform.baichuan-ai.com/", "apiLink": "https://platform.baichuan-ai.com/", "aiCoreShare": 100, "sourceUrls": [ "https://platform.baichuan-ai.com/", "https://www.baichuan-ai.com/" ], "versions": [ { "name": "Baichuan 4", "variants": [ "Baichuan4-Turbo", "Baichuan4-Air" ], "note": "Current model variants offered through the Baichuan open platform." }, { "name": "Baichuan 3" }, { "name": "Baichuan 2" } ], "versionsUpdated": "2026-09-02" }, { "id": "talkie", "name": "Talkie", "company": "MiniMax", "region": "china", "tagline": "MiniMax AI companion and character product", "description": "Talkie is MiniMax's AI companion and character product. It is cataloged as a consumer application surface, separate from the MiniMax model API.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 72, "competitors": [ "character-ai", "zeta", "Replika", "Anima AI" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.talkie-ai.com/", "apiLink": null, "aiCoreShare": 100, "scoringRationale": "Talkie is a real consumer companion product in MiniMax's portfolio, but product-specific active users and strategic platform leverage are not disclosed, so it remains below model/API/video surfaces.", "sourceUrls": [ "https://www.talkie-ai.com/", "https://www.minimax.io/", "https://ir.minimax.io/" ], "adoptionSignal": "MiniMax lists Talkie as a current global product, but its 300M+ user figure is aggregated across the company portfolio and is not assigned to Talkie.", "versions": [ { "name": "Talkie", "note": "Current MiniMax AI companion and character product." } ], "versionsUpdated": "2026-09-02" }, { "id": "kanana", "name": "Kanana", "company": "Kakao", "region": "korea", "tagline": "Kakao's open-weight Korean foundation model, moving to agents in KakaoTalk", "description": "Kanana is Kakao's Korean-language foundation model family, released with open weights. Kanana-2 adds tool-calling and stronger instruction-following, and Kakao is moving it into agent form inside KakaoTalk and Kanana Search during 2026. Kakao is also part of Korea's 'AI for Everyone' national assistant programme. Model-specific adoption is not disclosed, and KakaoTalk's user base is distribution rather than Kanana adoption.", "users": "Not disclosed", "userCount": 0, "releaseDate": "2024-06", "technicalImpact": 78, "competitors": [ "hyperclova-x", "qwen", "exaone", "solar-llm", "Llama 4", "Mistral", "DeepSeek V3", "Perplexity", "You.com" ], "features": [ "Korean language optimization", "Open-weight Kanana releases", "Agentic AI functions such as tool calling", "Closed beta consumer AI mate service", "Suite modules: Kanana 1.5, Kanana Nano, KoGPT", "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based" }, "link": "https://kanana.ai/", "apiLink": "https://developers.kakao.com", "aiCoreShare": 100, "adoptionSignal": "Kakao released Kanana-2 with tool-calling and improved instruction-following, and is rolling agent-based Kanana into KakaoTalk and Kanana Search during 2026. Kakao joined Korea's 'AI for Everyone' national assistant programme. No model-specific adoption figure is disclosed; KakaoTalk's user base is excluded as parent-platform scale.", "developmentSpend": "Not disclosed; corporate foundation-model program.", "capitalEfficiency": "Unproven: visible technical output, but no disclosed broad user adoption or revenue against development spend.", "scoringRationale": "Raised from 76 to 78. Kanana-2 is a current open-weight generation with tool-calling, and Kakao's participation in the national 'AI for Everyone' programme is a real distribution commitment — more than the benchmark-and-CBT evidence the previous 76 was based on. Held at 78 because adoption remains undisclosed and KakaoTalk reach is parent-platform scale, not Kanana adoption. Below the 84 sovereign tier (EXAONE, HyperCLOVA X, Solar) because Kanana has neither third-party frontier recognition nor a sovereign foundation-model mandate.", "sourceUrls": [ "https://www.kakaocorp.com/page/detail/11557?lang=ENG", "https://www.kakaocorp.com/page/detail/11654", "https://huggingface.co/collections/kakaocorp/kanana-15", "https://kanana.ai/", "https://www.koreatimes.co.kr/business/tech-science/20260120/kakao-updates-kanana-2-releases-4-open-source-models", "https://github.com/kakao/kanana-2/" ], "nativeName": "카나나", "versions": [ { "name": "Kanana-2", "released": "2026", "variants": [ "Tool calling", "Improved instruction following" ], "note": "Current generation. Agent form rolling into KakaoTalk and Kanana Search through 2026." }, { "name": "Kanana 1.5" }, { "name": "Kanana Nano" } ], "versionsUpdated": "2026-07-29" }, { "id": "exaone-discovery", "name": "EXAONE Discovery", "company": "LG AI Research", "region": "korea", "tagline": "EXAONE scientific discovery model effort", "description": "EXAONE Discovery is LG AI Research's materials and drug-discovery platform. LG demonstrated current partner deployments with LG Household & Health Care and GS Caltex at ICML 2026, including compound screening and an immersion-cooling fluid; platform customers and revenue remain undisclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 63, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.lgresearch.ai/exaone", "apiLink": null, "aiCoreShare": 96, "adoptionSignal": "Named 2026 partner work with LG Household & Health Care and GS Caltex, including screening 420,000 candidate compounds in one day; users and revenue are not disclosed.", "developmentSpend": "Not disclosed.", "capitalEfficiency": "Unproven.", "scoringRationale": "Held at 63: current named partner deployments establish real R&D use, but no external customer count, product revenue or population-scale deployment is disclosed.", "sourceUrls": [ "https://www.lgresearch.ai/news/view?seq=664", "https://www.lgresearch.ai/blog/view?seq=536" ], "versions": [ { "name": "EXAONE Discovery", "note": "Current materials and drug-discovery platform demonstrated at ICML 2026." } ], "versionsUpdated": "2026-09-02" }, { "id": "cogvideox", "name": "CogVideoX", "company": "Zhipu AI", "region": "china", "tagline": "Z.ai's CogVideoX video-generation model family", "description": "CogVideoX is Z.ai's video-generation model family for text, image and start/end-frame inputs. CogVideoX-3 adds audio, higher-resolution output and improved motion, stability and instruction following through the Z.ai developer API.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 72, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://docs.z.ai/guides/video/cogvideox-3", "apiLink": "https://docs.z.ai/api-reference/video/generate-video", "aiCoreShare": 100, "sourceUrls": [ "https://docs.z.ai/guides/video/cogvideox-3", "https://docs.z.ai/api-reference/video/generate-video" ], "scoringRationale": "CogVideoX remains a maintained video API, but the catalog records less product-specific traction than Kling, Hailuo or Vidu.", "versions": [ { "name": "CogVideoX-3", "variants": [ "Text-to-video", "Image-to-video", "Start/end frame", "Generated audio" ] } ], "versionsUpdated": "2026-09-02" }, { "id": "exaone-path", "name": "EXAONE Path", "company": "LG AI Research", "region": "korea", "tagline": "LG AI Research's open pathology foundation-model family", "description": "EXAONE Path is LG AI Research's pathology foundation-model family for whole-slide cancer analysis and multi-omics research. EXAONE Path 2.5 extends the open research line beyond pathology images to RNA, CNV, SNP and methylation data; clinical adoption is not disclosed.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 62, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://www.lgresearch.ai/exaone", "apiLink": null, "aiCoreShare": 96, "adoptionSignal": "Open research releases and benchmark evidence are current; LG describes Path 2.5 as multi-omics research infrastructure, while hospital or commercial deployment remains a future goal.", "developmentSpend": "Not disclosed.", "capitalEfficiency": "Unproven.", "scoringRationale": "Held at 62 under the regulated-niche evidence rule: open research and benchmark strength do not substitute for disclosed clinical deployment, scans, users or revenue.", "sourceUrls": [ "https://www.lgresearch.ai/news/view?seq=649", "https://www.lgresearch.ai/blog/view?seq=574" ], "versions": [ { "name": "EXAONE Path 2.5", "variants": [ "Pathology images", "Multi-omics" ] }, { "name": "EXAONE Path 2.0", "released": "2025-07" } ], "versionsUpdated": "2026-09-02" }, { "id": "autoglm", "name": "Z.ai Coding Plan", "company": "Zhipu AI", "region": "china", "tagline": "Z.ai coding and agent product surface", "description": "Z.ai Coding Plan is the subscription and integration surface for using current GLM models in agentic coding tools and IDEs. It supports more than 20 tools and currently centers on GLM-5.2 for long-horizon software work.", "users": "Not disclosed", "userCount": 0, "releaseDate": "", "technicalImpact": 74, "competitors": [ "GPT-5", "claude", "Gemini 2.5", "Grok 4" ], "features": [ "Foundation model capability", "API or platform access", "Enterprise deployment", "Multimodal or language specialization" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://z.ai/subscribe", "apiLink": "https://docs.z.ai/", "aiCoreShare": 100, "sourceUrls": [ "https://z.ai/subscribe", "https://z.ai/blog/glm-5.2" ], "scoringRationale": "Z.ai Coding Plan is a current paid coding surface, but remains below Claude Code, Cursor and stronger disclosed developer platforms until usage is public.", "versions": [ { "name": "GLM Coding Plan", "variants": [ "GLM-5.2", "20+ supported tools", "MCP bundle" ] } ], "versionsUpdated": "2026-09-02" }, { "id": "grok", "name": "Grok", "company": "xAI", "region": "global", "tagline": "Frontier AI model and assistant product from xAI", "description": "Grok is xAI's assistant and model family for chat, coding and agent workflows. The recorded flagship is Grok 4.7, with configurable reasoning and tool calling. Current events require enabled search tools; the base model is not a live feed. Image, video and voice services use separate endpoints. X-platform distribution is not counted as product-specific Grok adoption.", "users": "600M+ monthly reach across X and Grok apps (company-reported mixed distribution signal)", "userCount": 0, "adoptionSignal": "xAI reports roughly 600M monthly reach across X and Grok apps, Grok availability through web/mobile/X/API, US government product surfaces, and enterprise distribution through Amazon Bedrock and Databricks.", "releaseDate": "2023-11", "technicalImpact": 87, "competitors": [ "chatgpt", "claude", "gemini", "qwen", "deepseek" ], "features": [ "xAI assistant and model family", "Consumer chat, voice, and image modes", "API access", "Canonical Grok product family" ], "pricing": { "api": "Usage-based or enterprise pricing", "enterprise": "Contact sales" }, "link": "https://grok.com/", "apiLink": "https://docs.x.ai/developers/models", "aiCoreShare": 100, "developmentSpend": "$20B Series E; Colossus I and II cited by xAI as more than one million H100-equivalent GPUs.", "capitalEfficiency": "low", "scoringRationale": "Impact 87 reflects first-party frontier-model control, tool support and commercial API distribution. Mixed X/Grok reach is not an assistant active-user count. This version refresh neither validates a new audience figure nor increases the score; stronger product-specific adoption evidence is needed.", "sourceUrls": [ "https://grok.com/", "https://docs.x.ai/developers/models", "https://x.ai/news/grok-4", "https://x.ai/news/series-e", "https://x.ai/news/grok-amazon-bedrock", "https://x.ai/news/grok-databricks", "https://x.ai/news/government" ], "versions": [ { "name": "Grok 4.7", "note": "Flagship text/coding model with configurable reasoning. Enable search tools for current web/X information." }, { "name": "Grok 4.5", "released": "2026-07", "note": "Earlier generation." }, { "name": "Grok 4" }, { "name": "Grok 3" } ], "versionsUpdated": "2026-10-01", "descriptionKo": "Grok은 채팅·코딩·에이전트 작업을 위한 xAI의 어시스턴트·모델 제품군입니다. 확인된 대표 모델은 추론 강도 조절과 도구 호출을 지원하는 Grok 4.7입니다. 최신 사건은 검색 도구를 활성화해야 조회할 수 있으며 기본 모델이 실시간 피드는 아닙니다. 이미지·동영상·음성은 별도 엔드포인트를 사용합니다. X 플랫폼 노출을 Grok 고유 사용자 수로 계산하지 않습니다." }, { "id": "motif-12b", "name": "Motif", "company": "Motif Technologies", "region": "korea", "tagline": "Open-weight Korean LLM trained from scratch.", "description": "Motif-2-12.7B (Base / Instruct / Reasoning) is Motif Technologies' open-weight Korean LLM family, ranked the most performant Korean open-weight model by Artificial Analysis at release. Motif joined Korea's sovereign-AI foundation-model project in February 2026 (consortium with KAIST and Trillion Labs) with a roadmap to a 300B reasoning LLM, 310B VLM, and 320B VLA, and raised a KRW 24B Series B in May 2026. Commercial adoption is not yet disclosed.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Sovereign-AI project team (Feb 2026); Artificial Analysis top Korean open-weight ranking; KRW 24B Series B May 2026", "technicalImpact": 72, "competitors": [ "Solar", "EXAONE", "HyperCLOVA X", "Llama 4" ], "features": [ "Open-weight LLM", "Korean-first", "From-scratch training", "Consortium-backed" ], "pricing": { "free": "Open weights path", "pro": "API / partner", "enterprise": "Contact sales" }, "link": "https://motiftech.io", "apiLink": null, "aiCoreShare": 98, "sourceUrls": [ "https://www.koreaherald.com/article/10679368", "https://en.wowtale.net/2026/05/28/234131/" ], "versions": [ { "name": "Motif-2 12.7B" } ], "versionsUpdated": "2026-07-29" }, { "id": "trida-7b", "name": "Trida", "company": "Trillion Labs", "region": "korea", "tagline": "Korean diffusion-transformer LLM.", "description": "Trida-7B is Trillion Labs' diffusion-transformer LLM within the Tri series (0.5B through the open-sourced Tri-70B, Korea's largest open-sourced model at release, trained from scratch on 1.5T EN/KO/JA tokens with intermediate checkpoints released for research). Trillion Labs is a member of the Motif-led sovereign-AI consortium (Feb 2026) and announced industrial world models for AI factories built with NVIDIA Omniverse and Nemotron (June 2026). Commercial adoption is not disclosed.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "Tri-70B open-source release; Motif sovereign consortium member; NVIDIA industrial world-model program (Jun 2026)", "technicalImpact": 66, "competitors": [ "Mistral", "Solar", "Llama 4", "qwen" ], "features": [ "Diffusion transformer", "Korean-first", "Pre-seed", "Research-led" ], "pricing": { "free": "Limited", "pro": "API / partner", "enterprise": "Contact sales" }, "link": "https://huggingface.co/trillionlabs/Trida-7B", "apiLink": null, "aiCoreShare": 96, "sourceUrls": [ "https://huggingface.co/trillionlabs/Trida-7B", "https://huggingface.co/trillionlabs/Tri-70B-preview-SFT", "https://www.prnewswire.com/news-releases/trillion-labs-is-developing-industrial-world-models-for-ai-factories-built-with-nvidia-omniverse-and-nvidia-nemotron-302793862.html" ], "versions": [ { "name": "Trida-7B" } ], "versionsUpdated": "2026-07-29" }, { "id": "a-x", "name": "A.X", "company": "SK Telecom", "region": "korea", "tagline": "SK Telecom's open-weight Korean foundation-model series", "description": "A.X is SK Telecom's Korean foundation-model series. The current A.X K2 generation is an open-weight 688-billion-parameter model with lightweight, vision-language, and audio derivatives; SK Telecom reports planned or active use across manufacturing trials, defense deployment, enterprise agents, startups, and A. assistant features.", "users": "Not disclosed", "userCount": 0, "adoptionSignal": "SK Telecom names KG Steel and KONEC manufacturing trials, planned defense-unit deployment, A. and A. Biz integration, a lightweight SK hynix deployment, and API access for three startups and government programs.", "technicalImpact": 80, "aiCoreShare": 98, "competitors": [ "exaone", "solar-llm", "hyperclova-x", "motif-12b" ], "features": [ "688B-parameter Korean foundation model", "Open-weight language model", "Vision-language and audio derivatives", "Named domestic deployment and trial pipeline" ], "pricing": { "free": "N/A", "pro": "Not yet commercial", "enterprise": "Contact SK Telecom" }, "link": "https://huggingface.co/skt/A.X-K2", "apiLink": null, "sourceUrls": [ "https://www.sktelecom.com/en/press/press_detail.do?idx=1673", "https://huggingface.co/skt/A.X-K2", "https://news.sktelecom.com/en/2533" ], "scoringRationale": "Raised from 78 to 80 under the sovereign-model evidence rule: A.X K2 now has named domestic deployment and trial evidence plus an open-weight release, but no disclosed global developer adoption or production-scale usage that would justify the 84+ ecosystem tier.", "versions": [ { "name": "A.X K2", "released": "2026-07", "variants": [ "A.X K2 688B", "A.X K2 Light", "A.X K2 VL Light-Preview 21B", "A.X K2 ALM 22B", "A.X K2 Raon-Speech 21B" ], "note": "Open-weight current generation with language, vision-language, and audio derivatives." }, { "name": "A.X K1", "released": "2025", "variants": [ "519B parameters", "Teacher model for sub-70B distillation" ] }, { "name": "A.X 4.0" } ], "versionsUpdated": "2026-09-02" } ]