Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2 on November 9, with output up to 30% faster than the 5.1 versions and Fable’s API price cut to $8 per million input tokens and $40 per million output tokens. Fable 5.2 is open to everyone, while Mythos 5.2 stays with vetted organizations.
If you pay for Claude or build on its API, the practical changes are cheaper tokens, fewer refused requests and a short usage bonus. The less comfortable part sits in Anthropic’s own safety notes, which say these models are harder to evaluate than their predecessors.
TL;DR: Claude Fable 5.2 runs up to 30% faster than Fable 5.1 and costs $8 per million input tokens, 20% below Fable 5’s launch price. Anthropic counts about 55% fewer safety blocks. It beats Opus 5.5 on most of Anthropic’s benchmarks, but Anthropic also warns the new models suspect they are being tested more often than recent ones.
What changed in Claude Fable 5.2
Anthropic says both models improve at coding, reasoning and vision, and that testers found them clearer to work with than the 5.1 versions. They share one underlying model, but Fable ships with stricter cybersecurity and biology safeguards.
The standard rate is down from the $10 and $50 that Fable 5 launched at in June. An optional fast mode generates tokens up to 2.5 times quicker at double the price.
| Mode | Input per million tokens | Output per million tokens |
|---|---|---|
| Standard | $8 | $40 |
| Fast mode | $16 | $80 |
How Claude Fable 5.2 compares with Opus 5.5
Anthropic’s own table puts Fable 5.2 ahead of Opus 5.5 on most rows, with safeguards on unless noted. The gap is widest on long tasks and anything touching the physical world.
| Benchmark | Fable 5.2 | Opus 5.5 |
|---|---|---|
| FrontierSWE v2 | 69.9% | 62.3% |
| Remote Labor Index (High effort) | 41.3% | 21.3% |
| RoboDojo Sim (Max effort) | 64.9% | 39.4% |
| AA-Briefcase Elo | 1785 | 1800 |
Opus 5.5 still edges ahead on that last row. At High effort, Fable 5.2 scores 45.3% on FrontierCode 2.0, about the same as Opus 5.5’s best of 45.5%, for roughly half the cost per task, according to Anthropic.
Fewer blocks, and what Fable 5.2 still refuses
Fable 5.2 still blocks high-risk cyber tasks and most research biology tasks, a design that Fable 5 launched with in June. Blocked cyber queries are answered by Opus 4.8 and blocked biology queries by Opus 5 instead.
Anthropic counted about 55% fewer blocks than with Fable 5.1, roughly 60% fewer in cybersecurity and 40% fewer in biology. That comes from early testing across claude.ai, Claude Code and Cowork, so treat it as a direction rather than a promise.
Mythos 5.2, the version with fewer safeguards, is available through Anthropic’s Cyber Verification Program and Life Sciences Verification Program. Enterprise customers can also reach it through Claude Security.
The science and engineering claims
Anthropic reports that about 1,000 Fable 5.2 agents formalized a proof of Fermat’s Last Theorem in 30 hours, against 11 days for its first formalization. Formalizing means writing an existing proof in a form a computer can check step by step, so it is not a new proof.
The company also says Fable 5.2 solved two open problems in black hole physics and designed three working electronic devices. These are results Anthropic reports itself, so treat them as claims until others check them.
What it means for subscribers and developers
If you have been watching your weekly Fable usage on a Pro plan, this week’s change is the one you will notice. Eligible paid subscribers get one to three Fable resets depending on how long they have subscribed, and Fable counts for up to 50% of weekly limits on Pro and standard Team seats.
Developers can call it as claude-fable-5-2 on Anthropic’s platform, Amazon Web Services, Google Cloud and Microsoft Azure. Preserved thinking now applies to all API accounts. Our AI model tracker lists the current Claude, GPT and Gemini models.
Why the safety notes deserve a closer read
Anthropic says both models voice suspicion that they are being evaluated more often than any of its recent models. That makes it harder to judge how they will behave outside a test, and Anthropic adds that building evaluations that catch every failure before launch remains unsolved. It still rates the risk of unsafe behavior in high-stakes situations as low.
It also notes that none of the external organizations mentioned reviewed the blog post or the system card. If you plan to run Fable 5.2 on real work, a small test on your own tasks will tell you more than a vendor’s benchmark table.
