Anthropic still leads business adoption with 43.5 percent of American companies paying for it, but the lead has stopped widening at the pace it did in spring. OpenAI closes the distance quarter after quarter and is now growing faster than Anthropic in that segment. The split shows up at the model level, not at the brand level.
Key Takeaways
- In July, Anthropic reached 43.5 percent of the American businesses tracked, against 39.7 percent for OpenAI.
- Fable 5 accounts for only 6 percent of the tokens bought from Anthropic, while GPT-5.6 Sol takes 25 percent on the OpenAI side.
- xAI posted its steepest monthly rise since July 2025 and reached 4 percent of businesses.
Have an AI Sum Up This Article
ChatGPTAnthropic leads at 43.5 percent but gains ground more slowly
The figures come from Ramp, the corporate card and spend management company, which watches payments across more than 70,000 American businesses. On the July reading, 43.5 percent of them were paying for an Anthropic subscription or tokens, up 1.1 percentage points month over month, a number Ramp lays out in its August index of business AI adoption.
OpenAI sits behind at 39.7 percent, up only 0.23 points over the same month. That puts the ChatGPT maker below the growth of overall AI adoption, which moved from a little over 50 percent of Ramp customers in March to nearly 56 percent in July.
The trajectory reads better against spring. In May the gap stood at two points, Anthropic at 41 percent and OpenAI at 39 percent, a moment we covered when Ramp’s index showed Anthropic ahead on professional usage.
The flip itself dates back to early spring, when Anthropic moved ahead of OpenAI for the first time on this ground. Four months later the top spot holds, but it no longer expands on its own.
The scope deserves stating. Ramp only sees companies that run through its products, which leaves out large accounts served by American Express or an equivalent, and tilts toward technology firms. The reading describes a segment rather than the whole market.
Fable 5 accounts for 6 percent of Anthropic token sales
The model-level breakdown explains the move better than the market shares do. Over the past month, Fable 5 made up 6 percent of the tokens bought from Anthropic and 11.4 percent of the dollars spent there, despite being the company’s most expensive model by a wide margin.
The comparison with the flagship across the aisle is stark. GPT-5.6 Sol takes 25 percent of OpenAI tokens and 23 percent of spend. OpenAI’s top model is being picked four times more broadly inside its own base than Anthropic’s is inside hers.
Price carries part of the weight, at roughly 10 dollars per million tokens for Fable 5 against 5 dollars for GPT-5.6 Sol. In raw spend, Fable 5 still generated close to 75 percent of what GPT-5.6 Sol brought in during July, which describes a rarer usage that costs a great deal more per call. That price gap extends the sequence that opened when OpenAI and Anthropic started an API price war.
Ara Kharazian, lead economist at Ramp, ties the move to how the products land in practice and rates GPT-5.6 Sol as really good, increasingly the developer’s pick. He argued the opposite for Fable 5, which he says disappointed in both adoption and real-world application given its price and the data retention requirements attached to it.
That last point is no footnote for a corporate buyer, and it is part of why OpenAI pushed an abuse monitoring system that keeps no customer data. Compliance has become a sales argument on the same footing as a benchmark score.
Pricing remains the backdrop. A model sold at twice the rate has to prove itself on specific work, and July’s double gap, on unit price and on token volume, says that case has not yet landed with most buyers.
More articles on Horizon
- AI Writing Shows Up on One in Three Web Pages
- OpenAI Watches for Abuse Without Keeping Your Data
- Claude Designed Proteins That Worked in the Lab
What the current quarter can still overturn
Into the third quarter so far, OpenAI closes ground faster than Anthropic consolidates hers. Kharazian tempered it immediately, noting there is still a month left in the quarter and framing that stretch as an eternity at the sector’s current pace.
The third name on the sheet is changing status too. xAI gained 0.94 points to reach 4 percent of businesses, its steepest monthly climb since July 2025. The base stays small, but the slope is the sharpest in the table.
Model serving platforms moved up alongside, to 6.1 percent of AI-using businesses, up 0.2 points. They open access to open source models and to a slice of the models developed in China, outside the two large American providers.
Budget dispersion is the harshest number in the reading. In July the top 1 percent of businesses spent a median of 7,400 dollars per employee, the top 10 percent spent 650 dollars, and the median firm spent 11.95 dollars. Adoption is broad, spending is anything but.
For buyers, the practical consequence sits in the default choice. A model priced at twice the rate has to show a visible quality gap on the tasks actually being run, otherwise the decision lands on price and retention clauses. The concentration seen in public institutions, where ChatGPT takes 88 percent of AI spending in the US Congress, shows what an installed default produces.
The next monthly reading will show whether OpenAI closes the gap for good or whether July was a launch-cycle bump. Anthropic keeps a lead of nearly four points, which is comfortable enough, and worth very little if the direction reverses two months running.
Follow the story on Horizon.
