Best AI Model vs. Most Popular AI Chatbot
Anthropic builds more top-50 models than any other lab. ChatGPT still reaches about 7 times as many Americans as Claude.
The signal
- PewPew Research Center asked US adults which AI chatbots they use (fieldwork 17–23 Feb 2026). ChatGPT leads among all adults and in every age group.
- ArenaOn the 2 Oct 2026 snapshot, Anthropic has 14 models in the overall top 50, more than any other lab.
- Our data@mistralai/mistralai grew fastest among 13 AI SDKs on npm, 2,992% in a year; @anthropic-ai/sdk grew 2,047% (npm registry, measured by us).
Anthropic builds 14 of the Arena's top 50 models and OpenAI builds 5. But 44% of US adults use ChatGPT and 6% use Claude.
Left: each lab's share of the Arena text top 50 (overall), 2 Oct 2026. Right: share of US adults who use that lab's chatbot, Pew, Feb 2026. Both in %.
The two columns measure different things (model quality rank vs consumer reach) at different dates. They are on one % scale for comparison, not as one quantity. Copilot is left out because it runs mostly on another lab's models. Pew figures are survey toplines; check Pew's terms before commercial reuse.
Show the data (5 rows)
| Chatbot | Lab | Top-50 models | Share of top 50 | US adults using |
|---|---|---|---|---|
| ChatGPT | OpenAI | 5 | 10% | 44% |
| Claude | Anthropic | 14 | 28% | 6% |
| Gemini | 10 | 20% | 24% | |
| Meta AI | Meta | 4 | 8% | 14% |
| Grok | xAI | 1 | 2% | 8% |
The Mode read
Rankings of AI products count users. Rankings of AI models count quality. We put the two side by side to ask one question: does model quality predict reach?
On this evidence, no. Anthropic holds 28% of the Arena top 50 and reaches 6% of US adults. OpenAI holds 10% and reaches 44%. Google is the one lab strong on both: 20% of the top 50 and 24% of adults, through Android, Search and Workspace.
Quality earns a place on the shortlist. Distribution decides who gets used. ChatGPT reached consumers first and became the default name. Gemini uses surfaces Google already owns. Claude grows mostly through developers and companies, a channel that a consumer survey does not measure.
Benchmarks get you on the shortlist. Distribution gets you used.
What we would back: Founders who already own a channel (a community, a marketplace, an install base) and add AI to it.