Anthropic has released Claude Fable 5, its first public 'Mythos-class' model, immediately resetting the competitive landscape against OpenAI. The new model scored 161 on the Epoch Capabilities Index, surpassing GPT-5.5 Pro's score of 159.
The performance gap widens substantially in specialized benchmarks. On FrontierMath tier 4, an evaluation designed for extreme mathematical reasoning, Fable 5 scored 88 percent compared to GPT-5.5 Pro's 75 percent.
The most commercially significant lead, however, appears in real-world coding. On SWE-Bench Pro, which uses tasks pulled from actual GitHub repositories, Fable 5 achieved an 80.3 percent success rate. Anthropic's previous flagship, Opus 4.8, scored 69.2 percent, while GPT-5.5 Pro trailed at 58.6 percent.
Positioned above the Opus line, Fable 5 features a novel safety architecture designed to default to the less capable Opus 4.8 when facing restricted queries. The model is also optimized for long-term autonomous operations, targeting agentic use cases that require independent execution over extended periods. Pricing ranges from $10 to $50 per million tokens through June 2026. This 21-percentage-point gap in software engineering capability represents a crucial deciding factor for enterprise procurement in development workflows.