{"source":"AIMap","schema":"benchmark/v1","updated":"2026-10-22","benchmark":{"id":"osworld-2","name":"OSWorld 2.0","category":"agent-computer","unit":"%","higherIsBetter":true,"status":"active","sources":["epoch"],"primaryVariant":null,"url":"https://osworld-v2.xlang.ai/","year":2026,"desc":{"en":"Long-horizon successor to OSWorld (2026) with multi-application workflows.","zh-hans":"OSWorld 的长流程后继版本（2026），任务跨多个应用。"},"measures":{"en":"Binary task success (partial credit reported separately).","zh-hans":"任务完全成功率（部分得分另计）。"},"limits":{"en":"Few systems evaluated so far; expensive to run.","zh-hans":"目前评测过的系统还少，运行成本高。"},"contamination":"low","contaminationNote":{"en":"New tasks.","zh-hans":"新任务。"},"stats":{"results":14,"models":9,"latest":"2026-07-24","bySourceType":{"official":14}}},"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-5","modelName":"Claude Opus 5","provider":"Anthropic","rawName":"Claude Opus 5","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: max; effort: max","score":31.43,"unit":"%","rank":1,"rankComputedBySite":true,"date":"2026-07-24","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"},{"model":"claude-opus-5","modelName":"Claude Opus 5","provider":"Anthropic","rawName":"Claude Opus 5","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: xhigh; effort: xhigh","score":30.21,"unit":"%","rank":2,"rankComputedBySite":true,"date":"2026-07-24","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"},{"model":"claude-opus-5","modelName":"Claude Opus 5","provider":"Anthropic","rawName":"Claude Opus 5","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: high; effort: high","score":28.98,"unit":"%","rank":3,"rankComputedBySite":true,"date":"2026-07-24","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"},{"model":"gpt-5-6-sol","modelName":"GPT-5.6 Sol","provider":"OpenAI","rawName":"GPT-5.6 Sol","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: max; effort: max","score":27.34,"unit":"%","rank":4,"rankComputedBySite":true,"date":"2026-07-09","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"},{"model":"claude-opus-5","modelName":"Claude Opus 5","provider":"Anthropic","rawName":"Claude Opus 5","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: medium; effort: medium","score":25.33,"unit":"%","rank":5,"rankComputedBySite":true,"date":"2026-07-24","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"},{"model":"claude-opus-5","modelName":"Claude Opus 5","provider":"Anthropic","rawName":"Claude Opus 5","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: low; effort: low","score":22.32,"unit":"%","rank":6,"rankComputedBySite":true,"date":"2026-07-24","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"},{"model":"claude-opus-4-8","modelName":"Claude Opus 4.8","provider":"Anthropic","rawName":"Claude Opus 4.8","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: max","score":20.6,"unit":"%","rank":7,"rankComputedBySite":true,"date":"2026-05-28","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"},{"model":"claude-opus-4-7","modelName":"Claude Opus 4.7","provider":"Anthropic","rawName":"Claude Opus 4.7","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: max; effort: max","score":18.2,"unit":"%","rank":8,"rankComputedBySite":true,"date":"2026-04-16","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"},{"model":"gpt-5-5","modelName":"GPT-5.5","provider":"OpenAI","rawName":"GPT-5.5","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: xhigh; effort: xhigh","score":13,"unit":"%","rank":9,"rankComputedBySite":true,"date":"2026-04-23","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"},{"model":"claude-sonnet-4-6","modelName":"Claude Sonnet 4.6","provider":"Anthropic","rawName":"Claude Sonnet 4.6","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: medium; effort: medium","score":9.3,"unit":"%","rank":10,"rankComputedBySite":true,"date":"2026-02-17","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"},{"model":"claude-sonnet-4-6","modelName":"Claude Sonnet 4.6","provider":"Anthropic","rawName":"Claude Sonnet 4.6","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: max; effort: max","score":8.3,"unit":"%","rank":11,"rankComputedBySite":true,"date":"2026-02-17","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"},{"model":"minimax-m3","modelName":"MiniMax-M3","provider":"MiniMax","rawName":"MiniMax-M3","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: enabled","score":4.6,"unit":"%","rank":12,"rankComputedBySite":true,"date":"2026-06-01","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"},{"model":"kimi-k2-6","modelName":"Kimi K2.6","provider":"Moonshot AI","rawName":"Kimi K2.6","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: enabled","score":4.6,"unit":"%","rank":12,"rankComputedBySite":true,"date":"2026-04-20","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"},{"model":"qwen3-7-plus","modelName":"Qwen3.7-Plus","provider":"Alibaba (Qwen)","rawName":"Qwen3.7-Plus","variant":"leaderboard","setting":"Binary accuracy; 500 steps; reasoning: thinking","score":2.8,"unit":"%","rank":14,"rankComputedBySite":true,"date":"2026-06-02","sourceType":"official","source":"epoch","sourceUrl":"https://osworld-v2.xlang.ai/","collected":"2026-10-10"}]}