OSWorld
OSWorld benchmarks multimodal AI agents on their ability to complete open-ended, real-world computer-use tasks (operating GUIs across web, files, and applications) in live operating-system environments via screenshots and mouse/keyboard control, measured by execution-based task success rate.
What this benchmark measures
OSWorld benchmarks multimodal AI agents on their ability to complete open-ended, real-world computer-use tasks (operating GUIs across web, files, and applications) in live operating-system environments via screenshots and mouse/keyboard control, measured by execution-based task success rate.
Rows on this page are sourced from public benchmark artifacts, leaderboard exports, or source-linked model reports. Each row keeps benchmark version, source model name, and available run details attached to the score.
The metric shown here is task success rate. It should be interpreted within OSWorld, not compared as part of a site-wide ranking.
Frequently asked
What is OSWorld?
OSWorld benchmarks multimodal AI agents on their ability to complete open-ended, real-world computer-use tasks (operating GUIs across web, files, and applications) in live operating-system environments via screenshots and mouse/keyboard control, measured by execution-based task success rate. It is a agents benchmark measured by task success rate.
What does task success rate mean on OSWorld?
OSWorld reports task success rate (%); higher is better. Scores are shown only within OSWorld and are never averaged with other benchmarks.
What is the top reported OSWorld score?
Claude Fable 5 has the top reported score on OSWorld: 85.0% (task success rate).
Why do OSWorld scores differ across runs?
Harness, scaffold, reasoning effort, and prompt setup change results, so two runs of the same model can differ. evals.report keeps each score with its run context so the differences stay visible.
Does evals.report rank models across benchmarks?
No. OSWorld scores are shown within their own metric; evals.report never combines benchmarks into a composite ranking or a single "best model".