A reported OpenRouter sample of roughly 120,000 companies showed spending on OpenAI and Anthropic roughly even in September 2026; earlier that year, Anthropic accounted for about three-quarters of spending in the sample. A separate Ramp snapshot briefly put OpenAI ahead in mid-September, before Anthropic regained a narrow lead by that week’s end.
The figures describe different groups and observation windows. The API rate cards offer another view of the contest: for short context, GPT-5.6 Sol and Claude Opus 5.5 are each listed at $4 per million input tokens and $20 per million output tokens.
Two spending snapshots, two different samples
| Sample | Described scope | Period | Reported spending pattern |
| OpenRouter | Roughly 120,000 companies using OpenAI and Anthropic tools | September 2026; comparison with earlier in 2026 | Spending was roughly even in September; Anthropic accounted for about three-quarters earlier in the year. |
| Ramp | 70,000 customers, mainly high-growth technology companies; individual developers were excluded | Mid-September 2026, through the end of that week | OpenAI briefly led spending, then Anthropic regained a narrow lead by week’s end. |
The snapshots point to a shifting mix of enterprise spending, not a single, continuous ranking. Ramp’s sample was weighted toward high-growth companies, including some Fortune 500 companies. OpenRouter’s sample included AI-focused startups, more established technology companies, and large corporations.
What the API rate cards charge
A token is a small unit of text processed by a model. API bills count input tokens sent to the model separately from output tokens it generates, so the two rates below are separate charges.
| Provider and model | Condition | Input per million tokens | Output per million tokens |
| OpenAI GPT-5.6 Sol | Short context | $4 | $20 |
| Anthropic Claude Opus 5.5 | Listed rate | $4 | $20 |
Those matching rates make a narrow comparison straightforward: one million input tokens plus one million output tokens totals $24 for either model at the listed rates. That equal token volume does not mean the models complete the same work, use the same number of tokens for a task, or perform equally well.
The listed rates vary by model and billing condition. OpenAI lists GPT-5.6 Sol at $8 per million input tokens and $30 per million output tokens for long context; its short-context rates are promotional through at least November 21, 2026. GPT-5.6 Terra is listed at $2/$12, and GPT-5.6 Luna at $0.20/$1.20 per million input/output tokens. Anthropic lists Claude Sonnet 5.5 at $2/$10 and Claude Haiku 5.5 at $0.10/$0.50 for prompts up to 100,000 tokens. All of these figures are API rates in U.S. dollars, not subscription prices.
Why some companies changed their model mix
Cost and provider dependence shaped some reported business decisions. David Zhu, Reevo’s co-founder and CEO, said the company shifted from Anthropic models toward OpenAI models partly to reduce reliance on one provider as well as for cost reasons; he also said its preference could change.
David Hsu, Retool’s CEO, said his company used OpenAI models more often and estimated spending about 20% less on them than on Anthropic models. That figure describes Retool’s own estimate, not a standard rate comparison across providers.
Hsu also cited a reported 30-day user-data retention period for Anthropic’s Fable 5 as a concern for Retool. He said the company did not use Fable 5 because it could not guarantee the data-deletion terms it sought. The account concerns Fable 5 specifically.
Price cuts and a late-September rollout
OpenAI was reported to have cut GPT-5.6 Luna’s price by 80% and GPT-5.6 Terra’s by 20% shortly after their launch. Anthropic began rolling out its Claude 5.5 family in late September, with lower cost and greater efficiency presented among its advantages.
