Frontier Log

Claude Opus 5.5 vs GPT-6.1 Sol

Claude Opus 5.5GPT-6.1 Sol
Released
Input / 1M tokens$4$2
Cached input / 1M tokens$0.20$0.10
Output / 1M tokens$20$10
Context window1M1,050,000
Output limit128K128,000

Both labs report

Terminal-Bench 4.0

Claude Opus 5.566.4%xhigh effort; standard error ±2.6 points
GPT-6.1 Sol55.9%max effort

Terminal-Bench Science 0.1

Claude Opus 5.558.7%max effort; standard error ±3.5 to 5 points
GPT-6.1 Sol57.0%max effort

AutomationBench 1.0.6Private set

Claude Opus 5.540.0%Run by Zapier · without fallback models; set and release 1.0.6 named in the Claude Sonnet 5.5 system card, which prints the same figure
GPT-6.1 Sol36.1%Run by Zapier · max effort; AutomationBench 1.0.6

OSWorld 2.1Offline set, partial score

Claude Opus 5.587.2%82-task offline subset, no internet access; max effort; mean of 5 attempts
GPT-6.1 Sol78.0%max effort

DeepSWE v1.1

Claude Opus 5.574.2%average of 5 trials
GPT-6.1 Sol75.2%high effort

Reported for one model

Humanity's Last ExamNo tools, full set

Claude Opus 5.564.4%
GPT-6.1 SolNo data

Humanity's Last ExamWith tools, full set

Claude Opus 5.567.7%max effort
GPT-6.1 SolNo data

SWE-bench ProSet not stated

Claude Opus 5.589.9%adaptive thinking, max effort; average of 5 trials
GPT-6.1 SolNo data

OSWorld 2.1Full set, partial score

Claude Opus 5.581.8%max effort; mean of 5 runs; September 10, 2026 task files; the system card labels it OSWorld 2.0
GPT-6.1 SolNo dataReported for another setting: Offline set, partial scoreReported for another version: OSWorld 2.0 (v2026.08.08), Offline set, partial score

OSWorld 2.1Full set, strict pass rate

Claude Opus 5.548.7%max effort; mean of 5 runs; September 10, 2026 task files
GPT-6.1 SolNo dataReported for another setting: Offline set, partial scoreReported for another version: OSWorld 2.0 (v2026.08.08), Offline set, partial score

OSWorld 2.1Offline set, strict pass rate

Claude Opus 5.553.2%82-task offline subset, no internet access; max effort; mean of 5 attempts
GPT-6.1 SolNo dataReported for another setting: Offline set, partial scoreReported for another version: OSWorld 2.0 (v2026.08.08), Offline set, partial score

FrontierCode 1.1Main set

Claude Opus 5.554.6%Run by Cognition · medium effort; Claude Code harness
GPT-6.1 SolNo data

FrontierCode 1.1Extended set

Claude Opus 5.565.3%Run by Cognition · medium effort; Claude Code harness
GPT-6.1 SolNo data

ChartographyNo tools

Claude Opus 5.564.4%adaptive thinking, max effort; average of 5 runs; Gemini 3.5 Flash grader
GPT-6.1 SolNo data

ChartographyWith tools

Claude Opus 5.589.0%adaptive thinking, max effort
GPT-6.1 SolNo data

Agents' Last Exam v1Measure not stated

Claude Opus 5.5No data
GPT-6.1 Sol58.6%max effort; harness not stated

MMMU-ProVision setting

Claude Opus 5.5No data
GPT-6.1 Sol86.3%max effort

Related

All comparisons