Claude Fable 5.2 and Mythos 5.2: what Anthropic's new models are and who they're for
Anthropic has launched Claude Fable 5.2 and Mythos 5.2, its most powerful models for coding and science. What they improve, what they cost and which one you should use.
Anthropic, the company behind the Claude assistant, unveiled two new models on November 9: Claude Fable 5.2 and Claude Mythos 5.2. They’re the company’s most powerful models, built for the hardest technical work: coding, science and analysis.
The essentials
| What they are | Anthropic’s most capable models, for very complex technical tasks |
| Fable vs Mythos | They’re the same model. Fable has stronger safeguards and is available to everyone; Mythos is only for approved organizations |
| What improves | Coding, reasoning and vision (understanding images) |
| Speed | Up to 30% faster than version 5.1 |
| Developer pricing | $8 per million input tokens and $40 per million output tokens |
| Fast mode | Optional, up to 2.5x faster, at double the price |
Fable vs Mythos: what’s the difference?
Under the hood they’re identical. The difference is the safeguards:
- Fable has strong safeguards around cybersecurity and biology, so nobody can use it to cause harm. That’s why it’s open to all users.
- Mythos is the version without those limits. Only organizations doing cybersecurity or medical research can request it, through Anthropic’s verification programs.
If you’re a regular user, Fable is the one for you.
What Anthropic says it can do
According to Anthropic, during internal testing these models:
- Helped complete software migrations that were previously unfeasible.
- Designed three working electronic devices from scratch.
- Solved two open problems in black hole physics.
- Formalized the proof of Fermat’s Last Theorem in 30 hours, meaning they wrote it in a form a computer can verify.
Anthropic also says more than 90% of the code changes merged at the company are now written by Claude, with human oversight.
Note: all of this comes from the company itself. These are internal tests not yet reproduced by third parties.
How it compares
Anthropic published a benchmark table. One of the clearest results is from a real-world coding test that measures whether the model’s proposed code changes are good:
Fable 5.2 comes out on top in most tests, but not all. In an automated office tasks benchmark, for example, Google’s Gemini 4 Carbon scores higher (53.9% vs 51.4%).
Do I need it? Which one to use
Anthropic now has a family of models, from largest to smallest. Its own advice is to start with Opus 5.5 for everyday work and switch only if needed:
| Model | Best for |
|---|---|
| Fable 5.2 / Mythos 5.2 | Huge, open-ended technical projects: complex coding, research |
| Opus 5.5 | Demanding everyday work: coding, writing, analysis |
| Sonnet 5.5 | A balance of quality, speed and price |
| Haiku 5.5 | Fast, repetitive tasks: summaries, text classification |
Fable can also act as an advisor to smaller models (they consult it when stuck) or as an orchestrator that hands simple tasks out to them.
What about safety?
Anthropic says it tested the model thoroughly before launch, including new tests to see whether it would try to escape isolated test environments. Its conclusion is that the risk of the model acting against instructions remains low. Fable’s safeguards block dangerous uses as before, but with fewer unnecessary refusals.
If you’re just getting started, read what AI is and how a chatbot works and how to write good prompts. Wondering which assistant to choose? See ChatGPT vs Gemini vs Claude.
