Anthropic’s best AI model struggles to attract users as cheaper tools thrive
Simon Willison’s Weblog Simon Willison
Anthropic says its revenue run rate hit $65bn in July, but its priciest model is barely showing up in use. Cheap tools are doing the heavy lifting, which is awkward for a premium AI pitch.
Based on reporting by Simon Willison’s Weblog, Simon Willison — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic’s numbers are still climbing fast. According to people familiar with the matter cited by the FT, its annualized revenue for July reached $65bn, up from $47bn in May. The company also expects Q3 to be profitable, using the same model it used to call Q2 profitable. It told investors it now has 6,000 customers spending at least $100,000 a year.
But the model mix tells a different story. Ramp’s AI index, which looks at billing data from 70,000 companies using Ramp cards, suggests Anthropic’s customers are not flocking to its newest premium model. In July 2026, Opus 4.8 accounted for 28.0% of Anthropic model spend, while Sonnet 4.6 took 8.3% and Fable 5 took 8.0%.
Opus 5, meanwhile, showed up with just 3.5% of spend. That is not exactly a roaring debut, though the timing matters: Ramp’s breakdown covers July, and Opus 5 was only released on July 24. Still, the numbers fit the idea that Fable’s cost has made it a less popular choice than cheaper options.
OpenAI is also still growing, but at a different pace. The FT says its annualized revenue has risen 35% in the quarter to date to more than $40bn, helped by GPT 5.6 launching in July after a slower start to the year. The broader picture is familiar by now: AI companies can post eye-watering revenue run rates, while users quietly gravitate toward the cheaper stuff.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI that the hype cycle keeps trying to skip: customers like progress, but they love invoices less. If the best model is pricey enough to become the least-used model, that’s not a side note, it’s the business model talking. Closed-model bragging rights are nice; usage data is less polite.
Read more about this at: Simon Willison’s Weblog