Frontier Log

Benchmark · Active

OSWorld 2.1

Created by XLANG Lab, 16 Sep 2026 · Higher is better · Website

020406080Oct 202679.6%Kimi K3.1 Pro020406080Oct 202679.6%Kimi K3.1 Pro

Settings and run setups

Task setScoringHarnessProviders
Full set (108 tasks)Partial scoreNot statedAnthropicTencent
Full set (108 tasks)Partial scoreCommon internal frameworkMeta
Full set (108 tasks)Strict pass rateNot statedAnthropic
Full set (108 tasks)Strict pass rateCommon internal frameworkMeta
Offline set (82 tasks)Partial scoreNot statedAnthropic
Offline set (82 tasks)Partial scoreGemini CUA harnessGoogle
Offline set (82 tasks)Strict pass rateNot statedAnthropic
Offline set (82 tasks)Not statedNot statedOpenAI
OpenAI compares GPT-6.1 Terra with Claude Sonnet 5.5’s 83.9%, which Anthropic lists under “8.9.3 OSWorld 2.1 offline subset”
Not statedNot statedKimi Agent harnessMoonshot AI
Tencent lists Kimi K3.1 Pro’s 79.6 under “OSWorld 2.1 (Full, partial score)”

Download image

Results

ModelScoreSource
Kimi K3.1 Pro24 Oct 202679.6%Kimi K3.1 Pro Tech Blog: Advancing Open IntelligenceChecked 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.Kimi Agent harness · max reasoning effortOur capture
Tencent lists it under “OSWorld 2.1 (Full, partial score)”

All benchmarks