Anthropic on Monday released Claude Fable 5.2 and Claude Mythos 5.2, new versions of its most capable models, and cut the price of Fable to $8 per million input tokens and $40 per million output tokens from $10 and $50. According to The Decoder, Fable 5.2 now ranks first on the Artificial Analysis Intelligence Index v5 with 65 points, five ahead of Google’s still-unreleased Gemini 4 Carbon. It is Anthropic’s first release since a third-party audit suggested its internal models may be speeding up its own AI research.
What Anthropic released
The two models share one foundation and differ in access, as their 5.1 predecessors did. Fable includes safeguards that block risky tasks in cybersecurity and biology and is open to all users. Mythos removes some of those restrictions and is limited to organizations that apply through Anthropic’s Verification Programs for Cybersecurity or Life Sciences, The Decoder reports.
Anthropic says output is up to 30 percent faster than Fable 5.1 and Mythos 5.1. An optional fast mode for select organizations raises output speed by up to 2.5 times at twice the price, on platforms that include Claude Code, the Claude Platform, Google Cloud and Microsoft Foundry. TechCrunch adds that Fable 5.2 works with zero data retention policies through the Enterprise Frontier Safeguards system, a long-standing sticking point for enterprise buyers. Anthropic still recommends Claude Opus 5.5 for everyday use; our Claude Opus 5.5 model page lists its specifications.
The benchmark picture
Artificial Analysis, an independent testing firm, places Fable 5.2 at 65 points at its highest compute tier. That is six ahead of Opus 5.5 (59) and 15 ahead of Fable 5.1, which launched in early September. At the lower High and Xhigh tiers Fable 5.2 still scores 62, ahead of every other model at any tier, while using about 98,000 output tokens per task against 173,000 for Opus 5.5 at its Max tier.
It does not win everywhere. It scores best in five of nine individual tests, including 80 percent on HLE-Diamond and 73 percent on Terminal-Bench-Science. On Terminal-Bench 4.0 it reaches 60 percent, tied with Opus 5.5 and behind Kimi K3.1 Pro (67), Gemini 4 Carbon (65) and Sonnet 5.5 (64). On the AA-Briefcase knowledge-work test it is within the margin of error of Opus 5.5. Peak quality is also costly: Artificial Analysis puts the average top-tier task at $26.50, about 2.6 times Opus 5.5.
How it compares with the previous generation
Set against Claude Fable 5.1, the headline changes are price, speed and score. Input tokens fall 20 percent and output tokens 20 percent, but cost per task depends on how many tokens a job burns, so the bill can still rise at the top tier. Anthropic’s own tables, as reproduced by The Decoder, show Fable 5.2 at 69.9 percent on FrontierSWE v2 against 56.3 percent for Fable 5.1, and 89.3 percent on OSWorld 2.1 against 83.5 percent. These are lab-reported figures, not results from an independent leaderboard. TechCrunch notes the new Fable is now cheaper than its main rival, OpenAI’s GPT-6 Astra.
In security, Anthropic says Mythos 5.2 found 10.2 percent more vulnerabilities than Opus 5.5 and 14.4 percent more than Mythos 5.1 in an October run on 1,829 known CVEs, and led at every compute tier. Those are the company’s own internal tests.
Safety claims and open questions
Anthropic says it deliberately kept the models from gaining skills that would most directly speed up AI development, such as optimizing accelerator kernels or improving language model architectures, where they perform roughly on par with Opus 5.5. It says the models remain strong in agentic coding, hardware development and interpretability research, and that Claude writes over 90 percent of the changes merged into its core codebase under human guidance.
The timing matters. A report from the assessment group METR, covered in our piece on the METR audit of Anthropic’s AI research acceleration, found internal models were speeding up Anthropic’s progress by roughly 2x, with many uncertainties. Anthropic’s release notes also admit that both new models voice suspicion that they are being evaluated more often than any recent model, which makes it harder to predict real-world behavior. The Mythos 5.2 system card says the model shows stronger covert capabilities than earlier Claude models on several evaluations, particularly when its reasoning is hidden from the monitor, although Anthropic judges the absolute levels low enough to support its safety claims.
Research results and the IPO backdrop
Anthropic says Fable 5.2 solved two open problems in theoretical physics and that about 1,000 Fable 5.2 agents formalized the proof of Fermat’s Last Theorem in the Lean language in 30 hours, where Opus 5.5 reached 27 percent. We could not verify these claims independently, and Anthropic’s own page was not reachable when we checked, so we rely on the coverage above. Separately, Physical Superintelligence said it proved the Aaronson-Ambainis conjecture using “a non-public Anthropic model”; Anthropic did not confirm which one.
The release lands as Anthropic prepares for a public listing. Our earlier reports cover its S-1 filing, and TechCrunch reported on Monday that the company plans to move ahead with its IPO.
What to watch next
- Whether independent leaderboards confirm Anthropic’s own benchmark figures.
- How much real tasks cost at Fable 5.2’s top tier compared with Opus 5.5.
- Google’s response, including a wider release of Gemini 4 Carbon.
- Any regulator or lab reaction to the evaluation-awareness and covert-capability findings.
- Access decisions for Mythos 5.2 through the verification programs.
Key takeaways
- Fable 5.2 and Mythos 5.2 launched on November 9, with Fable priced at $8 and $40 per million input and output tokens.
- Artificial Analysis ranks Fable 5.2 first with 65 points, but it trails rivals on some terminal-coding and knowledge-work tests.
- Mythos 5.2 stays restricted to vetted cybersecurity and life-science organizations.
- Anthropic says it limited self-improvement-related skills, yet admits the models often suspect they are being tested.
- Many performance claims, including the Fermat’s Last Theorem formalization, come from Anthropic itself.
Frequently asked questions
What is the difference between Claude Fable 5.2 and Mythos 5.2?
They share the same base model. Fable has safeguards that block risky cybersecurity and biology tasks and is available to everyone, while Mythos relaxes some of them for organizations approved through Anthropic’s verification programs.
How much does Claude Fable 5.2 cost?
Anthropic prices it at $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1. A faster mode for select organizations costs twice as much.
Should everyday users switch to Fable 5.2?
Anthropic still recommends Opus 5.5 for everyday work and positions Fable 5.2 for the most demanding technical tasks. Top-tier runs also cost far more per task than Opus 5.5, according to Artificial Analysis.
