benchgap
Agentic · terminal

Terminal-Bench-Science 0.1 leaderboard

As of 2026-11-21, the highest measured score on Terminal-Bench-Science 0.1 is 77.1% by GPT-6.2 Astra. 40 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1GPT-6.2 Astra77.1%measured
2Claude Mythos 5.275.3%measured
3Claude Fable 5.274.6%measured
4GPT-6.2 Sol68.0%measured
5Gemini 4 Carbon66.3%estimated ± 10.7 pp, low confidence
6GPT-6 Astra64.6%measured
7Gemini 4 Neon60.7%estimated ± 10.7 pp, low confidence
8Claude Sonnet 5.559.9%measured
9Claude Opus 5.558.7%measured
10GPT-6.1 Terra58.3%measured
11Claude Mythos 5.157.9%estimated ± 14.2 pp, low confidence
12GPT-6.1 Sol57.0%measured
13Claude Fable 5.152.6%measured
14Gemini 4 Hydrogen51.4%measured
15Gemini 4 Argon51.3%estimated ± 10.7 pp, low confidence
16Claude Opus 547.7%estimated ± 10.7 pp, low confidence
17GPT-6 Sol42.7%estimated ± 10.7 pp, low confidence
18Pareto 26.942.4%estimated ± 14.2 pp, low confidence
19Pareto 26.10 Preview42.0%estimated ± 14.2 pp, low confidence
20Kimi K3.1 Pro38.1%measured
21Muse Wonder37.3%estimated ± 10.7 pp, low confidence
22Qwen4 Max37.1%estimated ± 14.2 pp, low confidence
23Claude Fable 533.8%estimated ± 10.7 pp, low confidence
24Ling 3.1 Flash31.5%estimated ± 14.2 pp, low confidence
25GLM-5.327.7%estimated ± 10.7 pp, low confidence
26Claude Haiku 5.525.7%estimated ± 10.7 pp, low confidence
27Grok 4.723.7%estimated ± 10.7 pp, low confidence
28DeepSeek V4.1 Flash 101723.2%estimated ± 10.7 pp, low confidence
29GPT-5.6 Sol23.2%estimated ± 10.7 pp, low confidence
30ERNIE 5.5 Flash Preview21.0%measured
31GLM-5.3-Flash20.8%estimated ± 10.7 pp, low confidence
32Hy419.8%measured
33LongCat-2.515.2%estimated ± 14.2 pp, low confidence
34MiniMax M3.1 Flash14.4%estimated ± 14.2 pp, low confidence
35Muse Ember10.2%estimated ± 10.7 pp, low confidence
36Step 5 Preview7.2%estimated ± 14.2 pp, low confidence
37Mistral Large 45.3%estimated ± 10.7 pp, low confidence
38MiMo-V2.6-Pro4.3%estimated ± 10.7 pp, low confidence
39SWE-22.4%estimated ± 14.2 pp, low confidence
40Inkling 2 Mini1.4%measured
41SWE-2 Flash0.4%estimated ± 14.2 pp, low confidence
42Claude Opus 4.80.0%estimated ± 10.7 pp, low confidence
43Claude Sonnet 50.0%estimated ± 10.7 pp, low confidence
44Ember-10.0%estimated ± 10.7 pp, low confidence
45Gemini 3.7 Flash0.0%estimated ± 10.7 pp, low confidence
46Gemini 3.8 Flash0.0%estimated ± 10.7 pp, low confidence
47GPT-5.6 Luna0.0%estimated ± 10.7 pp, low confidence
48GPT-5.6 Terra0.0%estimated ± 10.7 pp, low confidence
49GPT-6 Luna0.0%estimated ± 10.7 pp, low confidence
50Grok 4.50.0%estimated ± 10.7 pp, low confidence
51Grok 4.60.0%estimated ± 10.7 pp, low confidence
52Hy4 flash0.0%estimated ± 10.7 pp, low confidence
53Kimi K30.0%estimated ± 10.7 pp, low confidence
54MiMo-V2.6-Flash0.0%estimated ± 10.7 pp, low confidence
55Muse Spark 1.30.0%estimated ± 10.7 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General