ChatGPT app, GPT models and Codex in one AI family
Strength
92.23/3 rated
See which AI products are available, who makes them, and how the alternatives compare.
Start with an AI family or a specialist tool, then compare alternatives and the company behind each one. These are selected entry points into the catalog, not a ranking of every available product.
Each family brings its assistant, models and coding tool together. Open one page to choose how to use it.
ChatGPT app, GPT models and Codex in one AI family
Strength
92.23/3 rated
Claude app, Claude models and Claude Code in one AI family
Strength
91.23/3 rated
Google’s AI apps, models, coding tools and enterprise platform in one family
Strength
91.23/3 rated
Grok app, models and Grok Build in one AI family
Strength
703/3 rated
177 profiles · products & portfolios
379 profiles · products & portfolios
41 profiles · products & portfolios
Strength is the unweighted geometric mean of product value, adoption and durability, displayed on 0–100. ≈ marks editorial ratings; open a product to inspect the rationale, comparison and sources.
Map the ecosystem
Break down today's ecosystem: what each layer does, who operates there, and where the strengths and gaps are. Compare global, Chinese and Korean players across the current agentic-AI stack.
Start with the stack mapFollow the bigger picture
Now place that stack in the bigger story. Compute, models, agents, science and physical AI are overlapping waves of change. We are in Wave 3, with room to grow. Science and physical AI will need new stack maps of their own.
Compute & memory
The physical foundation
Frontier models
The capability race
Agentic AI
Now · still expanding
AI for science
Hypotheses meet experiments
Physical AI
A longer-term direction
2026-10-22 · Reviewed developments and editorial interpretation
Our inference: the useful moat question is whether customers keep a workflow when its underlying model changes. Harvey has disclosed organizational deployment; product retention and inference margins still need separate evidence.
Beam, Muse Spark Open and Step 5 Preview now have weight releases. Sparse models still need memory for all weights. Compare quality-adjusted throughput, serving reliability and licence terms before choosing a self-hosted route.
Timestral targets numerical forecasting on CPUs, while Tinker makes post-training accessible. These are candidates to test against current baselines. A narrow benchmark lead or a specialist label alone does not establish durable value.
Solar Pro 2 and Pro 3 are scheduled to end service on October 30 KST. Solar Pro 4 and Mini 4 are the current alternatives. Test tool calls and retries before migration; a pricing table does not tell you the cost of a completed task.
Every material score, evidence, listing, release, and eligibility change — dated and sourced.
Compare products side by side across adoption, versions, pricing, evidence, and Product Strength and its three components.
Explainable shortlists for Korean assistants, models, document AI, infrastructure, healthcare, and physical AI.
Three criteria, one Product Strength score, and the evidence behind it.
Product and company coverage, reviewed scores and the remaining research backlog.