claude fable 5.2: Anthropic Cuts Price to $8/$40 and Claims Coding Lead

Anthropic's new top model, claude fable 5.2, arrived on November 9 alongside Claude Mythos 5.2, and the headline is price as much as power: $8 per million input tokens and $40 per million output tokens, with outputs up to 30% faster than the 5.1 generation. Fable is open to all users, while Mythos, the same underlying model with fewer safeguards, stays limited to vetted cybersecurity and life-sciences organizations.

What actually happened

Anthropic's launch post says Fable and Mythos are "the same underlying model." Fable ships with strong cybersecurity and biology safeguards so it can be offered widely; Mythos goes to organizations approved through Anthropic's Life Sciences and Cyber Verification Programs. Developers can call it as claude-fable-5-2 on the Claude Platform, and the company says it is also available on Amazon Web Services, Google Cloud and Microsoft Azure.

CNBC reported that the new models cost 20% less than the 5.1 generation. Anthropic also announced a launch promotion: paid subscribers on eligible plans get one to three extra Fable usage resets this week, depending on how long they have subscribed.

The company is not telling everyone to switch. Its own guidance says users should start most coding and knowledge tasks with Claude Opus 5.5, its "strongest daily driver," and move up to Fable for the hardest open-ended work. Fable can also act as an "Advisor" that smaller models consult, or as an "Orchestrator" that hands simple jobs to Haiku, Sonnet or Opus. We covered those smaller models in claude opus 5.5 and claude haiku 5.5.

The benchmark claims

All of these numbers come from Anthropic's own evaluation table, so treat them as vendor claims until outside testers repeat them:

• FrontierSWE v2: 69.9% for Fable 5.2, versus 62.3% for Opus 5.5 and 65.5% for GPT-6 Astra.
• Humanity's Last Exam with tools: 74.2%, versus 57.2% listed for GPT-6 Astra.
• OSWorld 2.1 computer use (partial offline set): 89.3%, versus 87.2% for Opus 5.5.
• RoboDojo Sim, run internally: 64.9%, versus 27.6% for Fable 5.1.

Not every line is a win. On AutomationBench, a business-workflow test, Anthropic's table shows Gemini 4 Carbon at 53.9% ahead of Fable 5.2 at 51.4%. That is a useful reminder that one company's chart is built to tell its story, even when it is honest about the numbers.

The post also makes several big claims from internal testing: that the model helped formalize the proof of Fermat's Last Theorem in 30 hours, designed three working electronic devices and helped migrate a production service from Python to Rust that now starts about 26 times faster. We could not verify those independently.

Why the timing matters

The claude fable 5.2 launch arrives during Anthropic's IPO roadshow week. CNBC framed the launch as a product push ahead of a listing that could happen as soon as next week. It also matters for prediction markets that track which lab has the best public model, since a new flagship can change chat-arena rankings once enough votes come in. Rankings on those boards usually take days or weeks to settle for a new model.

The safety angle is part of the pitch too. Anthropic says it ran new evaluations for behaviors such as sandbox escapes and believes high-stakes misalignment risk remains low, with full details in the system card.

Price check

At $8 input and $40 output per million tokens, Fable 5.2 is still far above the small models built for high-volume agent work. Anthropic's pitch is that Fable at a lower effort setting can match Opus 5.5 at a higher one for less money per task, which only holds if your tasks look like its benchmarks.

What to do as a reader

If you build on Claude, read the official launch post and the system card before changing models, then test claude fable 5.2 on your own tasks at different effort levels. Anthropic itself says Medium effort is the default in Claude apps and that Fable is meant for hard problems, so for routine work the cheaper models may still be the better deal.

Keep an eye on independent benchmarks and arena rankings over the next few weeks. Vendor tables are a starting point, not a verdict.