Inference
EN
Terminal-Bench

Terminal-Bench 4.1.0

Agentic terminal task suite run by the Terminal-Bench team.

ModelsModels with a result.23
Top resultBest result on this benchmark.65.8%Claude Opus 5.5
Top-3 spreadPoints from first to third.4.9 pts
UpdatedDate the source changed the results.29 Sep 2026

Result and cost

22 of 23 results

Results

23 results
#ModelResultThe score from the source.CostUS dollars to run one task.TimeTime to run one task.
01Claude Opus 5.5max effort · Claude Code65.8% ±1.6$14.3080 min
02Claude Sonnet 5.5max effort · Claude Code63.9% ±1.5$22.222.1 h
03GPT-6.1 Solmax effort · Codex60.9% ±1.5$1.9253 min
04GPT-6 Astramax effort · Codex60.3% ±1.4$9.9047 min
05Claude Fable 5.1max effort · Claude Code57.3% ±2.0$18.9265 min
06Claude Opus 5max effort · Claude Code52.1% ±1.7$18.0980 min
07GPT-6 Solmax effort · Codex50.3% ±1.6$4.4044 min
08Claude Fable 5max effort · Claude Code44.9% ±2.0$22.0270 min
09GLM-5.3max effort · Claude Code41.2% ±1.7$8.271.6 h
10Grok 4.7xhigh effort · Grok Build38.8% ±1.8$11.161.6 h
11GPT-5.6 Solmax effort · Codex38.5% ±2.0$7.7040 min
12GLM-5.3-Flashnone effort · Claude Code35.5% ±1.8$0.361.7 h
13Qwen3.8 Max (0902)max effort · Claude Code27.0% ±2.0$14.052.8 h
14Claude Opus 4.8max effort · Claude Code23.6% ±1.8$19.6486 min
15GPT-5.6 Terramax effort · Codex21.5% ±1.7$5.2542 min
16Grok 4.6high effort · Grok Build19.7% ±1.6$10.8839 min
17Gemini 3.8 Flashhigh effort · mini-SWE-agent18.2% ±1.7$5.5433 min
18GPT-5.6 Lunamax effort · Codex17.6% ±1.5$1.0568 min
19GPT-6 Lunamax effort · Codex15.2% ±1.4$0.21938 min
20Muse Spark 1.3xhigh effort · Muse Code14.8% ±1.5$072 min
21Gemini 3.7 Flashhigh effort · mini-SWE-agent12.4% ±1.3$3.8228 min
21Grok 4.5high effort · Grok Build12.4% ±1.3$6.3553 min
23Claude Sonnet 5max effort · Claude Code12.1% ±1.5$29.101.8 h