Anthropic’s Claude Fable 5 just claimed the top spot across Code Arena: Frontend, Vision Arena, and Text Arena, leading its nearest competitor by 98 points. The model, which launched around June 9, is the company’s first publicly available entry in its Mythos-class lineup.
The performance gap isn’t subtle. On SWE-Bench Pro, a benchmark that tests AI models on real-world software engineering tasks, Fable 5 scored 80.3%, beating the next closest model by 11 points. It also topped Cognition’s FrontierCode benchmark at the Diamond level during medium reasoning efforts and excelled on ViBench, an evaluation focused on end-to-end app development.
What makes Fable 5 different
Fable 5 ships with a 1-million-token context window, allowing the model to ingest and reason over enormous codebases in a single pass. The pricing sits at $10 per million input tokens and $50 per million output tokens.
Early enterprise testing produced one particularly eye-catching result. Stripe reportedly completed multi-month engineering migrations in just one day using Fable 5, working across codebases containing up to 50 million lines of code.
