AI's Pricing Race Reverses as Frontier Models Get Pricier
After three years of falling token prices, the AI pricing race has flipped: GPT-5.5 and Claude Fable 5 launched at double their predecessors, while customers respond by optimizing usage and routing to cheaper challengers like DeepSeek. CNBC reports OpenAI spent ~$1.35 per dollar earned in 2025, even as Anthropic turned a Q2 profit on enterprise coding.
For three years the story of AI pricing was a race to the bottom, as OpenAI, Anthropic and their rivals slashed the price of a token to win developers. In 2026 that race has reversed. As CNBC reported on June 26, the frontier labs are now raising prices — and customers are responding not by paying up, but by learning to spend less.
The reason is arithmetic. According to figures cited by CNBC, OpenAI spent roughly $1.35 for every dollar it earned in 2025, and the cost of serving ChatGPT is projected at about $14 billion in 2026 against some $3.7 billion in revenue. That gap has forced a change in posture: GPT-5.5 launched at double the price of GPT-5.4, and Claude Fable 5 arrived at twice the rate of Opus 4.8. The era of subsidized tokens — what some developers had taken to calling "tokenmaxxing," throwing ever more context and reasoning at every request — is giving way to cost discipline.
Cheaper challengers are pulling in the opposite direction. DeepSeek's V4-Pro reportedly undercuts GPT-5.5 by more than 5× on input and 17× on output, giving budget-conscious teams a credible alternative for workloads that do not need a flagship model. The result is a market that now rewards efficiency: routing easy requests to small or open models, trimming prompts, caching aggressively and reserving frontier models for the tasks that actually require them.
The shift has also exposed who built a real business underneath the hype. Anthropic was tracking toward roughly $559 million in operating profit in the second quarter of 2026 at a reported $47 billion annualized revenue run rate, helped by a focused bet on enterprise coding, where Claude Code is said to hold about 40% of the generative-AI coding market. Owning a high-value, sticky use case has let it charge premium prices while still showing a profit — a contrast to rivals chasing scale.
Enterprises, meanwhile, are guarding their leverage. "If Gemini came up with something better … I want to be able to use that and not do a long-term commitment," MongoDB chief executive CJ Desai told CNBC, capturing a wider reluctance to lock into any single vendor at premium pricing. The message to the labs is blunt: keep the price worth paying, because switching has never been easier.
Want AI news before everyone else?
The morning's most important AI stories, straight to your inbox. No fluff.