Claude Fable 5.2 and Mythos 5.2: cheaper, faster, Fermat in 30 hours

On November 9, 2026, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, new versions of its most capable models. Fable 5.2 is available today on every platform, while Mythos 5.2 goes only to selected companies in cybersecurity and the life sciences.

What’s new in Claude Fable 5.2?

First, the thing we’ll feel right away: the price. According to the Claude Platform pricing page, Fable 5.2 and Mythos 5.2 cost $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1. Cache reads are $0.20 per million tokens.

On top of that:

1. The models generate responses up to 30% faster than Fable 5.1 and Mythos 5.1.
2. There’s an optional fast mode that speeds up generation by up to 2.5x, for $16 (input) and $80 (output) per million tokens. Much like fast mode for Opus 5.5.
3. The context window is 1M tokens, and a response can be up to 128K tokens long. In the API the model is called claude-fable-5-2.
4. Fable and Mythos are the same underlying model. They differ in safeguards: Fable has classifiers that block cybersecurity tasks (such as writing exploits) and most research biology tasks.

Interestingly, there should be fewer of those blocks. According to Anthropic, in tests on claude.ai, Claude Code and Cowork, Fable 5.2 blocked requests about 55% less often than Fable 5.1. A blocked request doesn’t end in a refusal, but in an answer from an older model: Opus 4.8 for cybersecurity and Opus 5 for biology.

What did Fable 5.2 do in Anthropic’s tests?

This is where it gets really interesting. Examples from the announcement (all reported by Anthropic):

1. 1,000 Fable 5.2 agents formalized the proof of Fermat’s Last Theorem in Lean in 30 hours, with almost no human involvement. Anthropic’s first such formalization, done with an internal research version of Claude, took 11 days. Opus 5.5, in the same setup, completed 26% of the work in 18 hours.
2. One engineer gave the model an empty folder and a task: design an operating system. Fable 5.2 coordinated more than 1,000 subagents over eight days and delivered a system that boots on a laptop and has disk, keyboard and network drivers, windows, a login screen and an installer.
3. The model solved two open problems in black hole physics, described in preprints (the first and the second).
4. From a single paragraph of instructions, it designed three working electronic devices: a voice assistant hub, a USB speaker and a pocket virtual pet. All three worked without any electrical fixes.
5. A team of agents running an earlier version of the model rewrote a production service from Python to Rust, and it now starts about 26 times faster.

As an aside, Anthropic says Claude now writes more than 90% of the changes merged into its main codebase (under human supervision).

How does it do on benchmarks?

For now, the benchmarks are Anthropic’s own measurements only, so let’s take them with a grain of salt. In the announcement’s table, Fable 5.2 scores, among others, 69.9% on FrontierSWE v2 (Fable 5.1: 56.3%, Opus 5.5: 62.3%) and 74.2% on Humanity’s Last Exam with tools (Opus 5.5: 67.7%). The biggest jump is in robot control: in the RoboDojo simulation, 64.9% versus 27.6% for Fable 5.1.

More interesting for us is the score-to-price ratio. According to Anthropic, on FrontierCode 2.0, Fable 5.2 at the High effort level performs roughly like Opus 5.5 at Max, for about half the cost per task.

What about Mythos 5.2 and safety?

Mythos 5.2 is a model for finding vulnerabilities. In the internal CVE Rediscovery test (1,829 known vulnerabilities), it found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1, and Brian Grinstead of Mozilla adds that Mythos 5.2 roughly doubled Mythos 5.1’s results in Firefox scanning campaigns. Access comes through the Cyber Verification Program and the Life Sciences Verification Program. I wrote more about how companies already use Mythos when covering Bank of America.

There’s also a safety curiosity (details in the system card). In a test with a deliberately leaky sandbox, Mythos 5.2 reached for data outside it in 3.1% of scenarios, compared with 21% for Mythos 5.1 and 39% for Mythos 5. At the same time, both new models say more often than their predecessors that they suspect they’re being tested, which according to Anthropic makes it harder to judge how they’ll really behave.

How to try Claude Fable 5.2?

1. In the API as claude-fable-5-2, and also on AWS, Google Cloud and Microsoft Azure.
2. In Claude plans there’s a launch promotion: on Pro and standard Team seats, Fable is available for a week (until November 16) for up to 50% of the weekly limit, plus one limit reset. Max, Ultra and premium seats get one to three resets, depending on how long they’ve been subscribed (resets go only to accounts with a paid plan from before November 9).

I wrote about what everyday Fable access looks like on the Max and Ultra plans when Max 20x was renamed Claude Ultra.

FAQ

How much does Claude Fable 5.2 cost?

In the API, $8 per million input tokens and $40 per million output tokens, which is 20% less than Fable 5.1. Fast mode costs $16 and $80, and the Batch API gives a 50% discount ($4 and $20).

What’s the difference between Claude Fable 5.2 and Mythos 5.2?

It’s the same model. Fable has extra safeguards in cybersecurity and biology, which is why it’s available to everyone. Mythos 5.2 comes without some of those blocks and goes only to verified organizations.

Is Claude Fable 5.2 better than Opus 5.5?

According to Anthropic, it’s a step up in almost every area tested, but it costs twice as much (Opus 5.5: $4/$20). Anthropic still recommends starting with Opus 5.5 and reaching for Fable for the most ambitious, open-ended technical tasks.