Deep research

Claude Fable 5 State-of-the-art on Cognition's FrontierCode eval, scoring highest among frontier models even at medium effort. More token-efficient than prior Claude models. Stripe reported a codebase-wide migration on a 50M-line Ruby codebase done in a day, versus an estimated two-plus months by hand. Highest score of any model on Hebbia's Finance Benchmark (senior-level reasoning), with major gains in document reasoning, chart and table interpretation, and problem solving. IMC reported near-across-the-board top results on trading-analysis evals (factual lookup, conceptual reasoning, root-cause analysis, expected-value analysis). New state-of-the-art for vision tasks. Extracts precise numbers from scientific figures and can rebuild a web app's source code from screenshots alone. Needs less scaffolding: beat Pokémon FireRed with a minimal vision-only harness, where earlier models needed complex helper harnesses. Stays focused across millions of tokens on long-running tasks and improves its outputs using its own notes. With persistent file-based memory in Slay the Spire, performance improved 3x more than Opus 4.8, and it reached the final act 3x more often. Works autonomously for longer than any prior Claude model