I Think Anthropic and OpenAI Have Found Product-Market Fit

In a post published on 27 May 2026, Simon Willison argues that OpenAI and Anthropic have found product-market fit with their coding agents, and that the clearest evidence is how both companies now charge their largest customers.

Enterprise customers now pay API prices

Willison pays $100 a month for Anthropic's Max plan and $100 for OpenAI's Pro plan. The ccusage tool estimated that the tokens he used in 30 days would have cost $1,199.79 on Claude Code and $980.37 on Codex: $2,180.16 of usage for $200. He assumed large companies got similar discounts, and found the opposite. Anthropic has moved its Enterprise plan to $20 per seat per month plus API pricing for usage; The Information reported the change on 14 April 2026, and a spokesperson said it dated from November 2025. OpenAI's Codex rate card says pricing changed on 2 April 2026 to align with API token usage instead of per-message pricing, and on 23 April that was extended to all existing ChatGPT Enterprise plans. As of April 2026 the enterprise price for Codex, Claude Code and Cowork matches the published API price, and the newest models cost more: GPT-5.5 is twice the API price of GPT-5.4.

Why he calls it product-market fit

ChatGPT is hugely popular but hard to monetise: in February OpenAI reported more than 900 million weekly active users and only 50 million paying consumer subscribers, 5.6% of them. Covering $1 trillion of infrastructure on $10 to $20 a month would require one to two billion subscribers staying four years. Companies spending $200 or more per user per month get there much faster, and coding agents burn far more tokens while becoming daily tools for well-paid professionals. The models released in November 2025 made agents genuinely useful, and six months later companies are spending real money. Of OpenAI's 703 open roles, Willison classifies 229 as enterprise sales and support, and 105 of Anthropic's 390.

Alarm stories, spending and a second inflection point

He started digging after a wave of stories about companies alarmed by their AI bills, and considers the most widely cited of them overblown. Uber's CTO said the company had maxed out its full-year AI budget only months into 2026, mostly because of Claude Code; since Claude Code only became really good in November, a budget set in 2025 failing to predict 2026 demand is unsurprising. Uber's COO noted that 25% of the company's code commits went through Claude Code last quarter, while saying it is hard to draw a line from that to more useful features shipped. The labs also spend heavily on inference: SpaceX's recent S-1 disclosed Cloud Services Agreements with Anthropic worth $1.25 billion a month through May 2029 for capacity across Colossus and Colossus II, and Anthropic said the deal would let it raise usage limits for Claude Code and the Claude API. Willison calls November 2025 an inflection point for capability and April 2026 one for revenue, and says the audited numbers in the coming IPO filings will show how real it is.