API Endpoints
For engineers and researchers who build AI systems
Endpoint quick-reference for Korean, global, and Chinese model providers: base URLs, OpenAI compatibility, and documentation links.
Quickstart
Most providers below speak the OpenAI chat/completions format — switching providers is usually just a base URL and API key swap.
# Any OpenAI-compatible provider: swap base_url and key
curl https://api.upstage.ai/v1/chat/completions \
-H "Authorization: Bearer $UPSTAGE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "solar-pro3",
"messages": [{"role": "user", "content": "안녕하세요!"}]
}'Korea
https://api.upstage.ai/v1Solar Pro models plus document parsing/OCR APIs. Point the OpenAI SDK at the base URL and it works.
https://clovastudio.stream.ntruss.com/v1/openaiHyperCLOVA X (HCX) models through NAVER Cloud. The /v1/openai path speaks the OpenAI chat/embeddings format with a CLOVA Studio API key.
Global
https://api.anthropic.com/v1/messagesNative Messages API with official SDKs. Claude 5 family: Opus 5, Sonnet 5, Fable 5, plus Haiku 4.5 for speed.
https://api.openai.com/v1GPT model family, Codex, image and speech APIs. The chat/completions format everyone else emulates.
https://generativelanguage.googleapis.com/v1betaGemini models with a generous free tier via AI Studio. OpenAI-compatible layer at /v1beta/openai/.
https://api.mistral.ai/v1Mistral model family with a free experimentation tier (La Plateforme).
https://api.x.ai/v1Grok models with live X data access, OpenAI- and Anthropic-compatible request formats.
https://openrouter.ai/api/v1One key, 400+ models across every major lab — the fastest way to compare models without separate accounts.
https://api.groq.com/openai/v1LPU-served open-weight models (Llama, Qwen, and others) at very high token speed, with a free tier for prototyping.
https://api.together.xyz/v1Serverless inference and fine-tuning across 200+ open-weight models.
China
https://api.deepseek.comDeepSeek chat and reasoner models at aggressively low prices — the API that reset the industry's cost expectations.
https://dashscope-intl.aliyuncs.com/compatible-mode/v1Qwen model family (text, vision, audio, image) via the international DashScope compatible-mode endpoint.
https://api.moonshot.ai/v1Kimi models including the open-weight K series, on the international platform endpoint.
https://api.z.ai/api/paas/v4GLM model family on the international endpoint, including a low-cost coding plan popular with agent builders.