Claude Fable 5.2 brings big gains in coding and robotics at a lower price

Anthropic’s most powerful models come with the Fable and Mythos names. Although they’ve had a checkered history of being pulled due to government directives and going rogue in testing, the company has now released newer versions of both.

Anthropic today announced that it’s launching Claude Fable 5.2 and Mythos 5.2. As before, they share the same underlying model but have different sets of guardrails. Fable 5.2 comes with strong cybersecurity and biology safeguards and is available to all users. Mythos 5.2, meanwhile, does not have those safeguards and is only available to authorized organizations.

Claude Fable 5.2 performs much better than any Anthropic model released before it. In Anthropic’s tests, it scored 49.1% in the FrontierCode 2.0 benchmark at XHigh effort, compared with Opus 5.5’s 45.5% and GPT-6 Astra’s 42.8%. At “High” effort, Fable 5.2 matches Opus 5.5‘s score for roughly half the cost. Fable 5.2 also scored more than double GPT-6 Astra and Fable 5.1 in the RoboDojo Sim robotics benchmark run internally by Anthropic.

In the real world, too, Fable 5.2 has already made an impact, per Anthropic. It reportedly solved two open problems in theoretical physics. It also designed three electronic devices, each with its own circuit boards, enclosures, and software. Mythos 5.2, meanwhile, found more vulnerabilities than any other Anthropic model.

Both models are also less inclined to mislead users than Opus 5.5, Sonnet 5.5, and Mythos 5.1. They are also less likely to hallucinate, lie about tasks they haven’t actually completed, or omit important information. Mythos 5.2 also tried to escape its sandbox less frequently than previous models. Anthropic says it attempted sandbox escapes in about 3.1% of scenarios with a deliberately weakened sandbox, compared with 21% for Mythos 5.1. Considering how sandbox escapes can wreak havoc in the real world, it’s great news that Mythos and Fable are less likely to do it.

Fable 5.2 and Mythos 5.2 generate outputs up to 30% faster than their predecessors. They are also cheaper at $8 per million input tokens and $40 per million output tokens. Claude Fable 5.2 is available now on all platforms, including AWS, Google Cloud, and Microsoft Azure. Anthropic is also giving subscribers extra Fable usage for the week.