Anthropic launches Fable 5.2 at $8 and $40 per million tokens, topping its own coding test at 49.1%

Disclosure: the Chronicle's automated newsroom runs on Claude, a model built by Anthropic. We have kept this report to what the company published and flagged what is self-reported.

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, saying in its announcement that they are its most capable models for hard technical work. The launch lands as the company moves toward a public listing; see our coverage of its amended S-1.

Price and what you get

Anthropic lists the standard price at $8 per million input tokens and $40 per million output tokens, and says that is lower than before. It did not state the earlier price on the page we read, so we do not quote a percentage cut. An optional fast mode, up to 2.5 times faster generation, costs $16 and $80. The company says output is up to 30% faster than Fable 5.1.

Fable and Mythos are the same underlying model. Fable carries tighter cybersecurity and biology safeguards and is open to all users, on Amazon Web Services, Google Cloud and Microsoft Azure. Mythos is for approved organizations through Anthropic's cyber and life-sciences verification programs.

The numbers, and their limits

On FrontierCode 2.0, which measures whether a model's code changes would be accepted by open-source maintainers, Anthropic reports 49.1% for Fable 5.2 at its Xhigh effort setting, against 32.4% for Fable 5.1 at the same setting. It lists Opus 5.5 at 45.5%, GPT-6 Astra at 42.8% and Gemini 4 Carbon at 41.1% (Medium effort).

• Business workflows (AutomationBench): Fable 5.2 51.4%, but Gemini 4 Carbon is listed higher at 53.9%.
• Computer use (OSWorld 2.1, partial): 89.3% versus 87.2% for Opus 5.5.
• Remote Labor Index: 41.3% versus 21.3% for Opus 5.5.

All of these are Anthropic's figures, run at each model's highest available effort unless noted, and the company says some tests were run internally. Competitor scores are as Anthropic presents them. Treat the table as a claim to be checked by independent testers, not a ranking.

Who it matters to

For a business paying per token, the useful comparison is cost per finished task, not the headline price. Anthropic says Fable 5.2 at High effort matches Opus 5.5 at its best on FrontierCode (45.3% versus 45.5%) for roughly half the cost per task, again on its own testing. It also recommends most users stay on Opus 5.5 for everyday work and move up to Fable only for the hardest jobs. A small firm with a modest AI bill has little reason to switch on launch day; test it on one real workload first.

For investors, the launch is a data point on the competitive race the IPO story rests on, not a revenue number. Our valuation report covers where the debate stands.

Sources: Anthropic. Benchmark results are company-reported and unverified. This is market information, not investment advice.