Frontier Log

Benchmark · Active

OSWorld 2.1

020406080100Oct 2025Jan 2026Apr 2026Jul 2026Oct 202615.7%Haiku 4.587.2%Opus 5.583.9%Sonnet 5.572.4%Haiku 5.5020406080100Jan 2026Jul 202615.7%Haiku 4.587.2%Opus 5.572.4%Haiku 5.5

Results

ModelScoreSourceSource check
Claude Haiku 5.57 Oct 2026 · Anthropic72.4%Introducing Claude Haiku 5.582-task offline subset, no internet access; max effort; mean of 5 attempts
1 other sourceThe same number is printed in 1 other lab document, linked below.
1
Claude Sonnet 5.528 Sep 2026 · Anthropic83.9%Introducing Claude Haiku 5.582-task offline subset, no internet access; max effort; mean of 5 attempts; reported later
1 other sourceThe same number is printed in 1 other lab document, linked below.
1
Claude Opus 5.522 Sep 2026 · Anthropic87.2%Claude Haiku 5.5 System Card82-task offline subset, no internet access; max effort; mean of 5 attempts; reported later
Source checkedA person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
Claude Haiku 4.515 Oct 2025 · Anthropic15.7%Introducing Claude Haiku 5.582-task offline subset, no internet access; fixed thinking budget, no server-side context compaction; mean of 5 attempts; reported later
1 other sourceThe same number is printed in 1 other lab document, linked below.
1

Benchmark

github.comHigher is better

XLANG LabIntroduced 16 Sep 2026

All benchmarks