Anthropic Claude Fable 5.2 and Mythos 5.2 Release Notes

Overview

Anthropic has introduced Claude Fable 5.2 and Claude Mythos 5.2, two new models designed to handle high-stakes, open-ended technical challenges. These models demonstrate significant improvements in coding, reasoning, and vision capabilities compared to their predecessors. While Fable 5.2 includes robust cybersecurity and biology safeguards for general use, Mythos 5.2 is available to verified organizations for specialized research in those fields.

Performance and Capabilities

Claude Fable 5.2 and Mythos 5.2 outperform the existing Opus 5.5 in most evaluated benchmarks, including agentic coding, scientific research, and hardware design.

Key Technical Achievements

Formalization: A team of 1,000 Fable 5.2 agents completed the formalization of Fermat’s Last Theorem in 30 hours with minimal human intervention, a task that previously took 11 days with a research version of Claude.
Physics Research: The models successfully solved two open problems in theoretical physics concerning black hole thermodynamics and quantum gravity.
Hardware Engineering: Fable 5.2 autonomously designed three working electronic devices, including circuit board routing and 3D-printed enclosure modeling, with minimal human guidance.
Robotics: The models show improved performance in controlling physical robots, outperforming specialized robotics models on the RoboDojo benchmark.

Operational Improvements

Anthropic has optimized the models for both speed and cost-efficiency.

Speed and Efficiency: Fable 5.2 and Mythos 5.2 generate outputs up to 30% faster than their 5.1 counterparts. An optional "fast mode" is available, increasing token generation speeds by up to 2.5x.
Pricing: The new models are priced at $8 per million input tokens and $40 per million output tokens. Fast mode is priced at $16 per million input tokens and $80 per million output tokens.
Safety and Anti-Distillation: The models feature enhanced protections against distillation attacks, with Anthropic reporting a 90% to 99% reduction in data harvesting by illicit distillers. Safety classifiers have also been refined, resulting in approximately 55% fewer false-positive blocks compared to Fable 5.1.

Community and Industry Reception

Early user feedback has highlighted the models' exceptional coding and agentic capabilities, though some users have expressed confusion regarding the versioning nomenclature (5.2 vs. 5.5) and the high cost of "xHigh" effort usage.

Insights from the Community

Workflow Integration: Users noted that Fable 5.2 excels when given complex, ambitious prompts, often outperforming Opus 5.5 in agentic tasks where the model is allowed to spend more tokens on reasoning.
Benchmarking Concerns: Some observers have questioned the methodology of comparing Fable 5.2 against other models at different effort levels, suggesting that performance gains may be tied to increased token expenditure rather than purely architectural intelligence.
Model "Nerfing": A recurring topic in community discussions is the perception that previous models (like Opus 5.5) are "nerfed" or throttled following the release of newer, higher-tier models, a claim that remains a point of contention among power users.

"I think Opus would do just as well if it spent more tokens. Maybe it's not a direct correlation with the model's intelligence, but rather with how effort is distributed." — Community observation on benchmark performance.

Availability

Claude Fable 5.2 is available across all major platforms, including AWS, Google Cloud, and Microsoft Azure. Organizations interested in the specialized capabilities of Mythos 5.2 can apply for access through Anthropic’s Life Sciences and Cyber Verification Programs.