Anthropic releases Claude Fable 5.2 and Mythos 5.2

Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2. As we reported when the 5.1 versions launched, Fable and Mythos are the same underlying model. Fable ships with cybersecurity and biology safeguards and is open to all users. Mythos is available only to approved organizations through Anthropic's Life Sciences and Cyber Verification Programs.

Pricing is $8 per million input tokens and $40 per million output tokens, which Anthropic says is lower than before. An optional fast mode generates tokens up to 2.5 times faster for $16 and $80. Anthropic says standard outputs are up to 30% faster than the 5.1 models.

The biggest gains in Anthropic's own benchmark table over Fable 5.1 are:

• Remote Labor Index: 41.3%, up from 17.9%
• RoboDojo Sim: 64.9%, up from 27.6% (run internally)
• Humanity's Last Exam (with tools): 74.2%, up from 65.6%
• FrontierSWE v2: 69.9%, compared with 62.3% for Opus 5.5 and 65.5% for GPT-6 Astra

Fable 5.2 did not win every test. Gemini 4 Carbon scored higher on AutomationBench, and Opus 5.5 had a higher Elo on AA-Briefcase.

In independent testing, Artificial Analysis scored Fable 5.2 at 65 on its Intelligence Index at Max effort, 62 at Xhigh and High, 56 at Medium and 54 at Low. Low effort was the cheapest at $3.10 per task, and cost per task varied by up to 8.6 times across effort levels. Low effort was also the fastest, at 6.5 minutes per task and 6.54 seconds to the first answer token. The model has a 1M-token context window, accepts text and image input, and outputs text.

Anthropic says that in prerelease testing the model solved two open problems in black-hole physics, designed three working electronic devices and formalized the proof of Fermat's Last Theorem in 30 hours. A team of agents running an earlier version moved a production service from Python to Rust, and the service now starts about 26 times faster. None of these claims has been independently verified.

On safety, Anthropic says the risk of high-stakes misalignment remains low and that it built new evaluations for behaviors such as sandbox escapes. It also says Fable's safeguards now block fewer harmless requests and that it has added its strongest protections yet against distillation attacks. For most tasks, Anthropic still recommends starting with Opus 5.5.