Add a benchmark

Benchmarks / Coding

Terminal-Bench 2.1 (Vals)

State-of-the-art set of difficult terminal-based tasks Scores are the values published by VALS AI. Source benchmark version 2.1.

72models
–Open Index weight
2026-09-28last updated
Vals AIpublisher

Terminal-Bench 2.1 (Vals)

Higher is better · best effort per model

Chart settings
72 of 72 selected

full table below

#ModelEffortScoreCost / taskHarness
1Opus 5.5high87.6%$0.496–
2GPT-6 Astramax87.3%$1.34–
3GPT-5.6 Solmax85.8%$1.02–
4Fable 5.1high85.0%$3.07–
5Opus 5high84.6%$0.887–
6GPT-6 Solmax83.1%$0.373–
7Sonnet 5.5high83.1%$0.625–
8Gemini 3.8 Flashhigh81.3%$1.54–
9Kimi K3max80.9%$0.341–
10Fable 5max80.5%$1.43–
11GPT-5.6 Lunamax79.0%$0.054–
12Muse Spark 1.3max79.0%$0.597–
13Grok 4.6high78.3%$0.454–
14Gemini 3.7 Flashhigh77.5%$1.02–
15GPT-5.6 Terramax77.5%$0.473–
16GPT-5.5high76.4%$0.742–
17MiMo V2.6 Flashhigh76.4%$0.023–
18DeepSeek V4.1 Flashhigh74.5%$0.099–
19Sonnet 5max74.5%$0.535–
20Gemini 3.5 Flashhigh74.2%$0.868–
21Gemini 3.6 Flashhigh73.8%$1.17–
22Grok 4.7xhigh73.4%$1.08–
23GPT-6 Lunamax73.0%$0.029–
24Opus 4.8max71.9%$2.41–
25GLM 5.3max71.5%$0.307–
26Gemini 3.1 Prohigh70.8%$0.583–
27Muse Spark 1.2xhigh69.7%$0.499–
28Muse Spark 1.1xhigh69.3%$0.417–
29Opus 4.7high68.5%$1.98–
30Grok 4.5high67.8%$0.284–
31GLM 5.2max67.8%$0.425–
32MiMo V2.6 Prohigh67.8%$0.073–
33Qwen 3.8 Maxmax67.4%$0.448–
34DeepSeek V4 Flash 0731max67.0%$0.021–
35Kimi K2.7 Codemax67.0%$0.257–
36GLM 5.3 Flashmax62.9%$0.028–
37Qwen 3.7 Maxmax61.0%$0.332–
38MiMo V2.5xhigh60.7%$0.022–
39Composer 2.5none58.4%–Cursor CLI
40Qwen 3.8 27Bxhigh58.4%$0.678–
41google/barium-bbhigh57.7%––
42Sonnet 4.6max57.3%$0.572–
43MiMo V2.5 Proxhigh57.3%$0.044–
44GLM 5.1max56.9%$0.232–
45Inkling Smallxhigh55.1%$0.128–
46Hy4–55.1%$0.140–
47GPT-5.4 Minihigh54.7%$0.287–
48DeepSeek V4 Pro 0813max54.7%$0.200–
49Gemini 3 Flashhigh53.9%$0.154–
50Kimi K2.6max53.6%$0.138–
51MiniMax M3max53.6%$0.206–
52Qwen 3.6 Plusmax53.2%$0.202–
53Qwen 3.7 Plushigh52.8%$0.096–
54Nemotron 3 Ultra 550B A55Bmax50.9%––
55Gemini 3.5 Flash-Litehigh50.2%$0.337–
56Ling 3.0 Flash–50.2%$0.067–
57Ling 3.0 Flash Fin–50.2%$0.043–
58DeepSeek V4max50.2%$0.239–
59MiniMax M2.7high48.7%$0.066–
60Inklingmax47.6%$0.631–
61Grok 4.20xhigh44.2%$0.407–
62Haiku 4.5xhigh43.8%$0.364–
63Grok 4.3high41.9%$0.982–
64Kimi K2.5max41.9%$0.086–
65GPT-5.4 Nanohigh41.6%$0.072–
66Mistral Medium 3.5high39.0%$1.19–
67Mercury 2.5high34.5%$0.215–
68Gemini 3.1 Flash-Litehigh34.1%$0.085–
69Laguna M.1–34.1%$0.301–
70Laguna XS.2–25.8%$0.254–
71Command Anone17.6%––
72Nemotron 3.5 Lightning–10.9%$0.014–