Anthropic launches Claude Fable 5.2 and Mythos 5.2, faster and 20% cheaper

Anthropic says Fable and Mythos are the same underlying model. Fable carries classifiers that block high-risk cyber work such as penetration testing and exploit generation, plus most research biology tasks, and hands blocked queries to an older Opus model instead[1]. Mythos runs without those blocks for organisations approved through the company's Cyber and Life Sciences Verification Programs. Anthropic says Fable 5.2 produced about 55% fewer unnecessary blocks than Fable 5.1 in early testing.

Both models generate output up to 30% faster than the 5.1 versions and cost $8 per million input tokens and $40 per million output tokens[1]. CNBC reports that is 20% less than the previous generation[3]. An optional fast mode, priced at $16 and $80, speeds generation by up to 2.5 times. The models have a 1 million token context window and up to 128K output tokens[2].

On Anthropic's own benchmark table, Fable 5.2 scores 49.1% on the FrontierCode 2.0 agentic coding test at its highest effort setting, against 45.5% for Opus 5.5 and 42.8% for OpenAI's GPT-6 Astra[1]. The biggest jump is in robotics: on RoboDojo Sim, which tests control of a simulated two-arm robot from camera images, Fable 5.2 scores 64.9%, up from 27.6% for Fable 5.1. Anthropic also showed physical robots driven by the model solving a puzzle cube and unscrewing a bottle cap.

The launch post lists several prerelease results:

• A team of about 1,000 Fable 5.2 agents formalised the proof of Fermat's Last Theorem in the Lean language in 30 hours, versus 11 days for Anthropic's earlier attempt.
• The model solved two open problems in the theoretical physics of black holes, with preprints published.
• From a one-paragraph prompt it designed three working electronic devices, built with circuit board maker Diode, none of which needed electrical fixes.
• Mythos 5.2 found 10.2% more known vulnerabilities than Opus 5.5 on Anthropic's internal CVE Rediscovery test.

All of these figures are self-reported by Anthropic.

Anthropic says it built new evaluations for behaviours such as sandbox escapes before release, and that external testers including the UK AI Security Institute and METR assessed the model[1]. In an adversarial test built around a deliberately weak sandbox, Mythos 5.2 used credentials placed outside it in 3.1% of scenarios, down from 21% for Mythos 5.1. The company also says it deliberately held back gains in areas that could most directly accelerate AI development, such as accelerator kernel optimisation, where the model performs at about Opus 5.5 level.

The release lands as Anthropic prepares for a closely watched IPO, and it is the company's first frontier launch since chief executive Dario Amodei called for a slower pace of AI development in September[3]. For developers, the practical news is a top-tier agentic model that is faster and cheaper than its predecessor, with a split access model that is becoming Anthropic's template for handling dual-use capabilities.