Frontier Log

Benchmark · InactiveSuperseded by OSWorld 2.1

OSWorld 2.0, release not stated

Created by XLANG Lab, 28 Jun 2026 · Higher is better · Website

020406080100Jul 2026Oct 202681.8%Opus 5.580.7%Fable 5.174.0%Opus 5020406080100Jul 2026Oct 202681.8%Opus 5.580.7%Fable 5.174.0%Opus 5

Settings and run setups

Task setScoringProviders
Not statedNot statedOpenAIAnthropic
Google lists GPT-5.6 Sol’s 62.6% under “OSWorld-2.0, Partial score, batch tool enabled”Google lists GPT-5.6 Terra’s 50.2% under “OSWorld-2.0, Partial score, batch tool enabled”
Not statedPartial scoreAlibaba
Not statedStrict pass rateAlibaba
Anthropic's Sep 10 filesPartial scoreAnthropic
Anthropic's Sep 10 filesStrict pass rateAnthropic

Download image

Results

ModelScoreSource
Claude Opus 5.522 Sep 202681.8%Claude Opus 5.5 System CardChecked A person checked this number against its source on 11 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.max effort · mean of 5 runs
Claude Fable 5.11 Sep 202680.7%Claude Opus 5.5 System CardChecked A person checked this number against its source on 11 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.max effort · mean of 5 runs · reported later
Claude Opus 524 Jul 202674.0%Claude Opus 5.5 System CardChecked A person checked this number against its source on 11 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.max effort · mean of 5 runs · reported later

All benchmarks