Research
Claude Opus 5.5 won back Anthropic’s lead in three weeks
In early September, OpenAI’s GPT-6 Astra pulled OpenAI level with Anthropic in what businesses and developers spend on the two labs’ AI models. Three weeks after Claude Opus 5.5 launched on September 22, Anthropic has won all of it back.
We track three public windows into that spending: OpenRouter, a marketplace developers use to call many models; Ramp’s AI Index, drawn from businesses’ connected billing; and Vercel’s AI Gateway, a developer platform. Each shows how spending splits across labs, and we compare Anthropic only with OpenAI.
All three trackers have swung back
- OpenRouter: OpenAI led on 14 days between September 9 and October 2. Since October 7, Anthropic has taken 65% to 71% of the two labs’ spending every day, above its August average of 65%.
- Ramp: Anthropic’s share rose from 51% in the week of September 14 to 62% in the week of October 5, close to its 63% before Astra. Its weekly spending grew 28% over those three weeks.
- Vercel: Anthropic’s share climbed from 53% on September 21 to 74% on October 14, back to where it stood when Astra launched, though still below its August level.
Opus 5.5 drove the swing
On Ramp, Opus 5.5 became the most-spent model of any lab in its third week, at more than three times GPT-6 Astra. It now accounts for 57% of the sample’s Anthropic spending. On OpenRouter, its daily spending quadrupled between its first full day and October 8 while its price per token held steady, so the growth came from usage, not pricing.
Much of that spending moved over from Anthropic’s older models: on Ramp, Claude Opus 5 fell 69% over the same three weeks. But Anthropic’s total still grew, so Opus 5.5 is also drawing new spending.
OpenAI’s newer models haven’t kept pace
GPT-6 Astra is fading. Its Ramp spending is down 37% from its September peak, and its average daily OpenRouter spending halved in the week of October 5.
OpenAI’s next releases haven’t replaced it. GPT-6 Sol, launched the same day as Opus 5.5, and GPT-6.1 Sol, released a week later, together reached about a fifth of Opus 5.5’s Ramp spending.
How we built this
This study draws on the same Village monitor as our September analysis. Our agent built it from OpenRouter’s published model spending, Ramp’s AI Index and Vercel’s AI Gateway leaderboards, and it refreshes all three every day.
Each tracker covers a different population with its own accounting, so we never pool them. Ramp has since revised its September figures: the week of September 14, which we previously reported as OpenAI’s first lead, now shows Anthropic narrowly ahead at 51%. OpenRouter stopped publishing dollar amounts on October 11, so its shares since then use its relative spending weights and its dollar figures end October 10. OpenRouter figures exclude a benchmark app that inflated Astra’s spending in mid-September.
Build your own signals on this data. Define what matters to your research and get alerted when it moves.
Proprietary signals from any source. Build pipelines with Village’s agent that gather data from the websites, APIs and documents your research depends on.