Anthropic upgrades Claude Fable and Mythos to 5.2: Faster, cheaper, better

What's new

Anthropic has upgraded Claude Fable and Claude Mythos to version 5.2, released on November 9, 2026. According to Anthropic, the models are better at coding, reasoning and image understanding.

Fable and Mythos are the same underlying model. Fable has strong safeguards against misuse in cybersecurity and biology and is available to everyone. Mythos is reserved for select organizations, which can apply for access through Anthropic's Life Sciences Verification Program and Cyber Verification Program. Companies can also get in through the Claude Security product.

Fable 5.2 is available in Claude apps and Claude Code and through the API under the name claude-fable-5-2. It can also be accessed through Amazon Web Services, Google Cloud and Microsoft Azure.

What's smart

The models write responses up to 30 percent faster than Fable 5.1 and Mythos 5.1. There is also an optional fast mode, which according to Anthropic increases speed by up to 2.5 times, but at a higher price.

On the default Medium setting, Fable 5.2 can solve hard tasks with fewer tokens than its predecessor. Tokens are the small chunks of text an AI model reads and writes in, and they are what you pay for.

According to Anthropic, Fable 5.2's safeguards against misuse are just as robust as before, but produce roughly 55 percent fewer unnecessary blocks. That should mean fewer rejections of perfectly ordinary questions.

Anthropic says that during testing, the models helped with software migrations, designed three working electronic devices and formalized the proof of Fermat's Last Theorem in 30 hours. This is the company's own information.

Cheaper or better?

The standard price is $8 per million input tokens and $40 per million output tokens. Anthropic calls that lower than for the 5.1 models, but the previous price is not stated in the source. Fast mode costs $16 and $80. A monthly price for regular users has not been disclosed, but paying subscribers are covered by a launch campaign with extra Fable usage.

On the FrontierCode 2.0 coding test, Fable 5.2 at High effort scored roughly the same as Opus 5.5 at Max effort, at about half the price per task. At Xhigh effort, Fable 5.2 scores 49.1 percent versus 45.5 for Opus 5.5 and 42.8 for OpenAI's GPT-6 Astra. On FrontierSWE v2, the figures are 69.9, 62.3 and 65.5 percent. On computer use (OSWorld 2.1), they are 89.3, 87.2 and 79.5 percent.

Mythos 5.2 scores 76.3 percent on the CyScenarioBench cybersecurity test versus 67.6 for Opus 5.5 and 61 for GPT-6 Astra. All figures are Anthropic's own, and the tests are based on different tasks for each lab. Prices in kroner or euros and availability in Denmark and the EU are not stated.

What it's good at

Fable 5.2 is aimed at coding, computer use, hardware design and robotics. On BenchCAD 2.1 (hardware development), it scores 83.5 versus 65.3 for Opus 5.5.

For most coding and knowledge tasks, however, Anthropic recommends starting with Opus 5.5 and then moving up to Fable or Mythos, or down to Sonnet 5.5 or Haiku 5.5. Fable 5.2 can also act as an advisor to smaller models or orchestrate them and hand them the simple tasks.

Anthropic assesses the risk of serious misalignment as low and has deliberately kept the model's AI research capabilities at Opus 5.5's level. The model has been tested by the UK AI Security Institute, Frontier Design and METR. Companies can get automatic security monitoring on their own infrastructure in a private beta, in which Bridgewater, Salesforce and Stripe are among the participants.