{"source":"AIMap","schema":"benchmark/v1","updated":"2026-10-22","benchmark":{"id":"osworld","name":"OSWorld (Verified)","category":"agent-computer","unit":"%","higherIsBetter":true,"status":"active","sources":["epoch"],"primaryVariant":null,"url":"https://os-world.github.io/","year":2024,"desc":{"en":"369 real computer tasks in a full Ubuntu desktop (browsers, office apps, file manager, terminal) checked by execution-based scripts.","zh-hans":"在完整 Ubuntu 桌面里完成 369 个真实电脑任务（浏览器、办公软件、文件管理、终端），由脚本按执行结果判分。"},"measures":{"en":"Task success rate of computer-use agents. Data compiled by Epoch AI from the OSWorld site.","zh-hans":"电脑操作 agent 的任务成功率。数据由 Epoch AI 从 OSWorld 官网汇编。"},"limits":{"en":"Step budgets (15/50/100) and agent frameworks differ between rows; compare like with like.","zh-hans":"各条目的步数上限（15/50/100）和 agent 框架不同，应同条件比较。"},"contamination":"medium","contaminationNote":{"en":"Task definitions are public.","zh-hans":"任务定义公开。"},"stats":{"results":20,"models":9,"latest":"2026-02-17","bySourceType":{"vendor":1,"official":19}}},"sources":[{"id":"epoch","name":"Epoch AI Benchmarking Hub","license":"CC BY 4.0","policy":"open","attribution":"Epoch AI, 'Capabilities & benchmarking', https://epoch.ai/benchmarks (CC BY 4.0)","site":"https://epoch.ai/benchmarks"}],"results":[{"model":"claude-opus-4-5","modelName":"Claude Opus 4.5","provider":"Anthropic","rawName":"Claude Opus 4.5","variant":"compiled","setting":"OSWorld; agent: Claude Opus 4.5; effort: no thinking","score":66.3,"unit":"%","rank":1,"rankComputedBySite":true,"date":"2025-11-25","sourceType":"vendor","source":"epoch","sourceUrl":"https://www.anthropic.com/news/claude-opus-4-5","collected":"2026-10-10"},{"model":"claude-sonnet-4-6","modelName":"Claude Sonnet 4.6","provider":"Anthropic","rawName":"Claude Sonnet 4.6","variant":"leaderboard","setting":"OSWorld; agent: claude-sonnet-4-6 (100 steps); effort: no thinking","score":72.1,"unit":"%","rank":1,"rankComputedBySite":true,"date":"2026-02-17","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"kimi-k2-5","modelName":"Kimi K2.5","provider":"Moonshot AI","rawName":"Kimi K2.5","variant":"leaderboard","setting":"OSWorld; agent: Kimi-K2.5","score":63.3,"unit":"%","rank":2,"rankComputedBySite":true,"date":"2026-02-03","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"claude-sonnet-4-5","modelName":"Claude Sonnet 4.5","provider":"Anthropic","rawName":"Claude Sonnet 4.5","variant":"leaderboard","setting":"OSWorld; agent: claude-sonnet-4-5-20250929 (100 steps); effort: no thinking","score":62.9,"unit":"%","rank":3,"rankComputedBySite":true,"date":"2025-10-30","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"claude-sonnet-4-5","modelName":"Claude Sonnet 4.5","provider":"Anthropic","rawName":"Claude Sonnet 4.5","variant":"leaderboard","setting":"OSWorld; agent: claude-sonnet-4-5-20250929 (50 steps); effort: no thinking","score":58.1,"unit":"%","rank":4,"rankComputedBySite":true,"date":"2025-10-30","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"claude-sonnet-4","modelName":"Claude Sonnet 4","provider":"Anthropic","rawName":"Claude Sonnet 4","variant":"leaderboard","setting":"OSWorld; agent: claude-4-sonnet-20250514 (50 steps)","score":43.9,"unit":"%","rank":5,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"claude-sonnet-4-5","modelName":"Claude Sonnet 4.5","provider":"Anthropic","rawName":"Claude Sonnet 4.5","variant":"leaderboard","setting":"OSWorld; agent: claude-sonnet-4-5-20250929 (15 steps); effort: no thinking","score":42.9,"unit":"%","rank":6,"rankComputedBySite":true,"date":"2025-10-30","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"claude-sonnet-4","modelName":"Claude Sonnet 4","provider":"Anthropic","rawName":"Claude Sonnet 4","variant":"leaderboard","setting":"OSWorld; agent: claude-4-sonnet-20250514 (100 steps)","score":41.4,"unit":"%","rank":7,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"claude-3-7-sonnet","modelName":"Claude 3.7 Sonnet","provider":"Anthropic","rawName":"Claude 3.7 Sonnet","variant":"leaderboard","setting":"OSWorld; agent: claude-3-7-sonnet-20250219 (50 steps)","score":35.8,"unit":"%","rank":8,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"claude-3-7-sonnet","modelName":"Claude 3.7 Sonnet","provider":"Anthropic","rawName":"Claude 3.7 Sonnet","variant":"leaderboard","setting":"OSWorld; agent: claude-3-7-sonnet-20250219 (100 steps)","score":35.6,"unit":"%","rank":9,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"computer-use-preview","modelName":"computer-use-preview-2025-03-11","provider":null,"rawName":"computer-use-preview-2025-03-11","variant":"leaderboard","setting":"OSWorld; agent: computer-use-preview (50 steps)","score":31.3,"unit":"%","rank":10,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"claude-sonnet-4","modelName":"Claude Sonnet 4","provider":"Anthropic","rawName":"Claude Sonnet 4","variant":"leaderboard","setting":"OSWorld; agent: claude-4-sonnet-20250514 (15 steps)","score":31.2,"unit":"%","rank":11,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"computer-use-preview","modelName":"computer-use-preview-2025-03-11","provider":null,"rawName":"computer-use-preview-2025-03-11","variant":"leaderboard","setting":"OSWorld; agent: computer-use-preview (100 steps)","score":30.5,"unit":"%","rank":12,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"claude-3-7-sonnet","modelName":"Claude 3.7 Sonnet","provider":"Anthropic","rawName":"Claude 3.7 Sonnet","variant":"leaderboard","setting":"OSWorld; agent: claude-3-7-sonnet-20250219 (15 steps)","score":27.1,"unit":"%","rank":13,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"computer-use-preview","modelName":"computer-use-preview-2025-03-11","provider":null,"rawName":"computer-use-preview-2025-03-11","variant":"leaderboard","setting":"OSWorld; agent: computer-use-preview (15 steps)","score":26,"unit":"%","rank":14,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"o3","modelName":"o3","provider":"OpenAI","rawName":"o3","variant":"leaderboard","setting":"OSWorld; agent: o3 (100 steps); effort: medium","score":23,"unit":"%","rank":15,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"o3","modelName":"o3","provider":"OpenAI","rawName":"o3","variant":"leaderboard","setting":"OSWorld; agent: o3 (50 steps); effort: medium","score":17.2,"unit":"%","rank":16,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"o3","modelName":"o3","provider":"OpenAI","rawName":"o3","variant":"leaderboard","setting":"OSWorld; agent: o3 (15 steps); effort: medium","score":9.1,"unit":"%","rank":17,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"qwen2-5-72b","modelName":"Qwen2.5-72B","provider":"Alibaba (Qwen)","rawName":"Qwen2.5-72B","variant":"leaderboard","setting":"OSWorld; agent: qwen2.5-vl-72b-instruct (100 steps)","score":5,"unit":"%","rank":18,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"},{"model":"qwen2-5-72b","modelName":"Qwen2.5-72B","provider":"Alibaba (Qwen)","rawName":"Qwen2.5-72B","variant":"leaderboard","setting":"OSWorld; agent: qwen2.5-vl-72b-instruct (15 steps)","score":4.4,"unit":"%","rank":19,"rankComputedBySite":true,"date":"2025-07-27","sourceType":"official","source":"epoch","sourceUrl":"https://os-world.github.io/","collected":"2026-10-10"}]}