Composite-Bench

Composite-Bench, a long-horizon browser-based computer-use benchmark built from deterministic enterprise-web environments, scores AI agents on pass@1 against certified optimal solutions across five headline slices and a sixth number-partitioning slice with a separate reward ladder. Claude Fable 5 ranks first with a score of 74 on the top imported result.

Long-horizon browser-based computer-use benchmark built from deterministic enterprise-web environments. Its five headline slices score pass@1 against certified optimal solutions, while a sixth number-partitioning slice reports a separate reward ladder for deliberately intractable tasks. Category: Agentic. Imported rows: 8. Top imported result: Claude Fable 5, rank 1, 74.