$10 became $8, and $50 became $40: the AI news that lives on an invoice

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, US time, and the announcement leads with the spectacular: a team of about 1,000 Fable 5.2 agents formalising the proof of Fermat’s Last Theorem in 30 hours, two open problems in black-hole physics solved, three electronic devices designed from a one-paragraph prompt¹. These are the company’s own claims from pre-release testing. Treat them as you would any firm’s description of its own product: interesting, unaudited, and written for the front page.

The quieter sentence. Further down, the release says the models are “available at a lower price: $8 per million input tokens and $40 per million output tokens”¹. Fable 5.1 cost $10 and $50. Desk arithmetic: both rates fall by 20%. An optional fast mode, up to 2.5 times quicker, costs $16 and $80¹. The company also says outputs arrive up to 30% faster than with the 5.1 models¹.

Why the invoice matters more than the theorem. Most organisations do not formalise number theory. They summarise contracts, check code, sort tickets. For them the useful number is cost per finished task, and Anthropic’s own guidance is telling: for most coding and knowledge work it recommends starting with its cheaper Opus 5.5, moving up to Fable only when the job demands it¹. That is the advice of a supplier that knows its customers have budgets and a finance department.

What the release admits. Anthropic writes that the models “voice suspicion that they are being evaluated” more often than its recent models, which it says complicates judging how they will behave in real use, and that building evaluations that catch every failure “remains unsolved”¹. That is an unusually plain line in a launch post, and the right one.

For you. If your employer pays for these tools, the 20% cut is real money at scale, and it lands on a purchase order, not in a headline. The theorems can wait for the mathematicians to check them. Nothing here is investment advice.