MMODELYST
Evals/τ²-bench
EVL

τ²-bench

benchmarkAgentssuccess %
Leaderboard · 100 models
success % — higher is better
1GLM-5.2 (max)
99.1
2JT-35B-Flash
99.1
3GLM-4.7-Flash
98.8
4GLM 5V Turbo
98.5
5GLM-5-Turbo
98.5
6Step 3.7 Flash
98.5
7Claude Fable 5
98.5
8GLM-5
98.2
9Qwen3.6 Plus
97.7
10GLM-5.1
97.7
11Grok 4.3
97.7
12Grok 4.20 0309
96.5
13DeepSeek V4 Pro
96.2
14Qwen3.6 Max Preview
95.9
15GLM-4.7
95.9
16Kimi K2.5
95.9
17Kimi K2.6
95.9
18Qwen3.5 397B A17B
95.6
19Gemini 3.1 Pro Preview
95.6
20MiniMax-M2.5
95.3
21Qwen3.6 35B A3B
95.3
22Gemini 3.5 Flash
95.3
23MiMo-V2-Pro
95
24MiMo-V2-Flash
95
25DeepSeek V4 Flash
95
26Qwen3.7 Max
94.7
27Step 3.5 Flash
94.4
28Claude Opus 4.8
94.4
29Qwen3.6 27B
94.2
30MiMo-V2.5-Pro
94.2
31Mistral Medium 3.5
94.2
32Qwen3.5 27B
93.9
33GPT-5.5
93.9
34Qwen3.5 122B A10B
93.6
35Grok 4.1 Fast
93.3
36Tri-21B-think Preview
93.3
37MiMo-V2-Flash (Feb 2026)
93.3
38Kimi K2 Thinking
93
39Grok 4.20 0309 v2
93
40Qwen3.7 Plus
93
41JT-MINI
93
42Hy3-preview
92.7
43Nova 2.0 Pro Preview
92.7
44Ring-2.6-1T
92.4
45Claude Opus 4.6
92.1
46GPT-5.2 Codex
92.1
47Qwen3.5 4B
92.1
48Muse Spark
91.5
49MiMo-V2-Omni
91.2
50DeepSeek V3.2
90.6
51MiMo-V2.5
90.6
52Grok 3 mini Reasoning
90.4
53Trinity Large Thinking
90.1
54Kimi K2.7 Code
90.1
55Ling-2.6-1T
89.8
56Claude Opus 4.5
89.5
57KAT Coder Pro V2
89.5
58Qwen3.5 35B A3B
89.2
59MiniMax-M3
88.9
60KAT-Coder-Pro V1
88.6
61Claude Opus 4.7
88.6
62Qwen3.5 Omni Plus
88.3
63MiMo-V2-Omni-0327
88
64MiniCPM-V 4.6 1.3B
87.7
65HyperCLOVA X SEED Think (32B)
87.4
66Step 3.5 Flash 2603
87.4
67Gemini 3 Pro Preview
87.1
68GPT-5.4
87.1
69MiniMax-M2
86.8
70GPT-5 Codex
86.8
71Qwen3.5 9B
86.8
72Mi:dm K 2.5 Pro
86.5
73Solar Pro 3
86.3
74GPT-5.6 Terra (max)
86.3
75Ling 2.6 Flash
86
76GPT-5.3 Codex
86
77MiniMax-M2.1
85.4
78GPT-5.6 Sol (max)
85.1
79GPT-5
84.8
80GPT-5.2
84.8
81MiniMax-M2.7
84.8
82GPT-5.6 Sol
84.8
83Qwen3.5 Omni Flash
84.5
84ERNIE 5.0 Thinking Preview
83.9
85Qwen3 Max Thinking (Preview)
83.6
86Qwen3 Max Thinking
83.6
87Nemotron 3 Ultra 550B A55B
83.3
88GPT-5.4 mini
83.3
89GPT-5.1 Codex
83
90GPT-5.1
81.9
91Nex-N2-Pro
81.6
92MiniCPM5-1B
81
93Tri-21B-Think
81
94Command A+
80.7
95o3
80.7
96Gemini 3 Flash Preview
80.4
97Nova 2.0 Omni
80.4
98GPT-5.6 Terra
80.4
99Qwen3 Coder Next
79.5
100LongCat Flash Lite
79.5
Benchmark health
saturating

Top models cluster near the ceiling — differences here are getting too small to mean much. Weight newer, harder benchmarks more.

98.5
top-10 mean
99.1
best
100
models
Recent changes
GPT-5.6 Terra (max)+86.36d ago
GPT-5.6 Terra+80.46d ago
GPT-5.6 Sol (max)+85.16d ago
GPT-5.6 Sol+84.86d ago
GPT-5.5 Instant (May 2026)+49.413d ago
What improves this score

Datasets & environments shown to raise it.

No data yet.