On 9 November 2026, Anthropic released two models: Claude Fable 5.2 and Claude Mythos 5.2. They are the same model, but Fable is publicly available with strong cybersecurity and biology safeguards, while Mythos requires an access request through special verification programmes. This post covers Fable. Not tried yet.
What’s new
- Speed: Anthropic says it generates output up to 30% faster than Fable 5.1.
- Price: $8 per million input tokens and $40 per million output tokens. An optional fast mode speeds up token generation by up to 2.5 times, at $16 and $80 respectively.
- Blocking: In Anthropic’s early evaluations on claude.ai, Claude Code and Cowork, blocks caused by safety classifiers fell by about 55% compared with Fable 5.1.
- Efficiency: On the FrontierCode 2.0 coding benchmark, Fable 5.2 at High nearly matched Opus 5.5 at its highest setting, at roughly half the cost per task, according to Anthropic.
Anthropic also describes a team of 1,000 Fable 5.2 agents that converted a proof of Fermat’s Last Theorem into a computer-verifiable proof in 30 hours. Other examples include an experimental operating system and three electronic devices designed from scratch. These are results Anthropic itself announced.
Why it’s worth a look
Anthropic’s strongest publicly available model is now cheaper and faster. Tasks that were very expensive on Fable 5.1, such as reviewing a large codebase or doing lengthy research, may now be affordable.
Still, Anthropic itself recommends starting with Opus 5.5 for most coding and knowledge work, and moving to Fable when the task is large and open-ended. Fable can also act as an advisor: Haiku, Sonnet or Opus works on the task and asks Fable about difficult problems. Advisor in Claude Code is experimental and available only through the Anthropic API and Claude Platform on AWS.
How to start
In Claude apps, choose Fable 5.2 from the model menu if your plan includes it. In Claude Code (v2.1.321 or later, with an account that has access to the model), you can run it directly using Anthropic’s model ID:
claude --model claude-fable-5-2To decide whether it suits your work, give it a task you’ve already completed with your current model and compare:
Here is a task I previously completed with another model: [describe the task and files]. Do it from scratch, then write at the end what you did differently and where you are uncertain about the result.
Compare quality, time and token count together.
Limits
- All performance figures come from Anthropic.
- Fable blocks high-risk cybersecurity activities, such as penetration testing and writing exploits, and most biology research tasks. Anthropic says the answer then comes from Opus 4.8 for cybersecurity and Opus 5 for biology, but in Claude Code this depends on the settings and model availability, and the request may be refused.
- In Claude apps, it requires an eligible plan. On some plans, its use counts against usage credits rather than plan limits. It is also available through the API with separate billing. During launch week, Anthropic gives eligible subscribers extra Fable usage.
- The page says nothing specific about Arabic support.