MMODELYST
Evals/Terminal-Bench Hard
EVL

Terminal-Bench Hard

benchmarkAgentssuccess %
Leaderboard · 100 models
success % — higher is better
1GPT-5.6 Sol (max)
65.9
2Claude Fable 5
62.9
3GPT-5.6 Terra
62.9
4GPT-5.6 Sol
61.4
5GPT-5.5
60.6
6Claude Opus 4.8
58.3
7GPT-5.4
57.6
8GPT-5.6 Terra (max)
57.6
9Gemini 3.1 Pro Preview
53.8
10Claude Sonnet 4.6
53
11GPT-5.3 Codex
53
12GPT-5.4 mini
52.3
13Claude Opus 4.7
51.5
14Qwen3.7 Max
50.8
15GLM-5.2 (max)
50.8
16KAT Coder Pro V2
49.2
17Claude Opus 4.5
47
18GPT-5.2
47
19Qwen3.7 Plus
47
20Claude Opus 4.6
46.2
21DeepSeek V4 Pro
46.2
22GPT-5.1
45.5
23Muse Spark
45.5
24Kimi K2.7 Code
44.7
25Qwen3.6 Max Preview
43.9
26Qwen3.6 Plus
43.9
27Kimi K2.6
43.9
28GLM-5
43.2
29GLM-5.1
43.2
30MiMo-V2.5-Pro
43.2
31GPT-5.5 Instant (May 2026)
42.4
32MiniMax-M3
42.4
33GPT-5.4 nano
42.4
34Gemini 3 Pro Preview
41.7
35MiMo-V2.5
41.7
36MiMo-V2-Pro
40.9
37Grok 4.20 0309
40.9
38Qwen3.5 397B A17B
40.9
39Gemini 3.5 Flash
40.9
40MiniMax-M2.7
39.4
41Gemini 3 Flash Preview
38.6
42Grok 4
37.9
43Grok 4.20 0309 v2
37.9
44GPT-5 Codex
37.9
45Grok 4.3
37.9
46GPT-5.2 Codex
37.1
47o3
37.1
48Nemotron 3 Ultra 550B A55B
36.4
49Gemma 4 31B
36.4
50DeepSeek V3.2
35.6
51Claude 4.5 Sonnet
35.6
52MiMo-V2-Omni-0327
35.6
53Step 3.7 Flash
35.6
54DeepSeek V4 Flash
35.6
55Kimi K2.5
34.8
56MiniMax-M2.5
34.8
57DeepSeek V3.2 Speciale
34.8
58GPT-5.1 Codex
34.8
59Qwen3.6 35B A3B
34.8
60Qwen3.6 27B
34.8
61MiMo-V2-Omni
34.8
62Claude 4.1 Opus
34.3
63Hy3-preview
34.1
64GPT-5.1 Codex mini
33.3
65GLM-5-Turbo
33.3
66Mistral Medium 3.5
33.3
67Qwen3.5 27B
32.6
68GPT-5
32.6
69GLM 5V Turbo
32.6
70Step 3.5 Flash 2603
32.6
71GLM-4.7
31.8
72Kimi K2 Thinking
31.1
73DeepSeek V3.2 Exp
31.1
74Claude 4 Opus
31.1
75Claude 4 Sonnet
31.1
76Ling-2.6-1T
31.1
77Qwen3.5 122B A10B
31.1
78North Mini Code
31.1
79MiMo-V2-Flash (Feb 2026)
31.1
80DeepSeek V3.1 Terminus
30.3
81MiniMax-M2.1
28.8
82GPT-5 mini
28.8
83Ring-2.6-1T
28.8
84JT-35B-Flash
28.8
85NVIDIA Nemotron 3 Super 120B A12B
28.8
86MiMo-V2-Flash
28
87Step 3.5 Flash
27.3
88Claude 4.5 Haiku
27.3
89Doubao Seed Code
26.5
90Qwen3.5 35B A3B
26.5
91Mercury 2
26.5
92Gemini 2.5 Pro
26.5
93MiniMax-M2
25.8
94GLM-4.6
25
95Command A+
25
96ERNIE 5.0 Thinking Preview
25
97Qwen3 Max Thinking
24.2
98Grok 4.1 Fast
24.2
99DeepSeek V3.1
24.2
100Qwen3.5 9B
24.2
Benchmark health
headroom

Plenty of room at the top — this benchmark still separates frontier models clearly.

59.4
top-10 mean
65.9
best
100
models
Recent changes
GPT-5.6 Terra (max)+57.66d ago
GPT-5.6 Terra+62.96d ago
GPT-5.6 Sol (max)+65.96d ago
GPT-5.6 Sol+61.46d ago
GPT-5.5 Instant (May 2026)+42.413d ago
What improves this score

Datasets & environments shown to raise it.

No data yet.