benchgap
Coding

VulcanBench v3 leaderboard

As of 2026-11-19, the highest measured score on VulcanBench v3 is 89.9% by Grok 4.5. 96 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Grok 4.589.9%measured
2Claude Fable 589.5%measured
3DeepSeek V4 Flash 073188.4%measured
4Claude Fable 5.287.6%estimated ± 3.9 pp, low confidence
5Gemini 4 Carbon87.6%estimated ± 3.9 pp, low confidence
6GPT-6.1 Sol87.6%estimated ± 3.9 pp, low confidence
7GPT-6.2 Astra87.6%estimated ± 3.9 pp, low confidence
8GPT-6.2 Sol87.6%estimated ± 3.9 pp, low confidence
9Claude Fable 5.187.6%estimated ± 3.9 pp, low confidence
10Muse Wonder Full87.6%estimated ± 3.9 pp, medium confidence
11Gemini 4 Argon87.6%estimated ± 3.9 pp, medium confidence
12DeepSeek V4.1 Flash 101787.6%estimated ± 3.9 pp, medium confidence
13Kimi K3.1 Pro87.6%estimated ± 3.9 pp, medium confidence
14Gemini 4 Hydrogen87.6%estimated ± 3.9 pp, medium confidence
15Muse Spark 1.387.6%estimated ± 3.9 pp, medium confidence
16Muse Wonder87.6%estimated ± 3.9 pp, medium confidence
17Grok 4.787.6%estimated ± 3.9 pp, medium confidence
18GPT-6 Astra87.1%estimated ± 0.7 pp, low confidence
19Claude Opus 5.587.1%estimated ± 0.7 pp, medium confidence
20Claude Sonnet 5.587.1%estimated ± 0.7 pp, medium confidence
21Claude Opus 587.0%measured
22GPT-5.6 Sol87.0%measured
23GPT-5.6 Terra87.0%measured
24Grok 4.687.0%measured
25Muse Spark 1.287.0%measured
26Gemini 3.7 Flash86.3%estimated ± 5.6 pp, low confidence
27Gemini 3.1 Pro86.2%estimated ± 5.6 pp, low confidence
28GPT-5.2-Codex85.8%estimated ± 5.6 pp, low confidence
29GPT-5.6 Luna85.5%measured
30Gemini 3.5 Flash85.4%estimated ± 5.6 pp, low confidence
31DeepSeek V4 Pro85.4%estimated ± 5.6 pp, low confidence
32GPT-5.3 Codex85.2%estimated ± 5.6 pp, low confidence
33Qwen3.7 Max85.1%estimated ± 5.6 pp, low confidence
34Kimi K2.684.9%estimated ± 5.6 pp, low confidence
35GPT-5.184.6%estimated ± 5.6 pp, low confidence
36Gemini 3 Pro84.6%estimated ± 5.6 pp, low confidence
37Nemotron 3 Ultra84.3%estimated ± 5.6 pp, low confidence
38Qwen3.6 Plus84.3%estimated ± 5.6 pp, low confidence
39GPT-584.2%estimated ± 5.6 pp, low confidence
40Inkling-Small84.2%estimated ± 5.6 pp, low confidence
41Muse Spark 1.184.2%estimated ± 5.6 pp, low confidence
42Gemini 3 Flash84.0%estimated ± 5.6 pp, low confidence
43GPT-5.1-Codex84.0%estimated ± 5.6 pp, low confidence
44Inkling83.9%estimated ± 5.6 pp, low confidence
45GPT-5.283.9%estimated ± 5.6 pp, low confidence
46Claude Opus 4.783.7%estimated ± 5.6 pp, low confidence
47Claude Opus 4.6 (Adaptive)83.4%estimated ± 5.6 pp, low confidence
48Grok 4.383.3%estimated ± 5.6 pp, low confidence
49Grok 4.2083.2%estimated ± 5.6 pp, low confidence
50GPT-5.483.0%estimated ± 5.6 pp, low confidence
51GPT-5.4 nano83.0%estimated ± 5.6 pp, low confidence
52Ling 3.0 Flash83.0%estimated ± 5.6 pp, low confidence
53o382.9%estimated ± 5.6 pp, low confidence
54Claude Opus 4.5 Thinking82.8%estimated ± 5.6 pp, low confidence
55GPT-5.1-Codex-Max82.7%estimated ± 5.6 pp, low confidence
56Claude Opus 4.882.7%estimated ± 5.1 pp, low confidence
57Qwen3.8-27B82.6%measured
58Qwen3.5 Flash82.6%estimated ± 5.6 pp, low confidence
59Grok 482.5%estimated ± 5.6 pp, low confidence
60Claude Sonnet 582.1%estimated ± 5.1 pp, low confidence
61GLM-4.782.0%estimated ± 5.6 pp, low confidence
62MiniMax M382.0%estimated ± 5.6 pp, low confidence
63o4-mini82.0%estimated ± 5.6 pp, low confidence
64Claude Sonnet 4.681.9%estimated ± 5.6 pp, low confidence
65Kimi K2.7 Code81.9%estimated ± 5.6 pp, low confidence
66GPT-5.4 mini81.6%estimated ± 5.6 pp, low confidence
67MiMo-V2.581.6%estimated ± 5.6 pp, low confidence
68GLM-5.181.6%estimated ± 5.6 pp, low confidence
69MiMo-V2.5-Pro81.6%estimated ± 5.6 pp, low confidence
70GLM-4.681.4%estimated ± 5.6 pp, low confidence
71Qwen3.8 Max81.2%measured
72Grok 4.1 Fast (Reasoning)81.2%estimated ± 5.6 pp, low confidence
73Gemini 3.1 Flash-Lite80.9%estimated ± 5.6 pp, low confidence
74MiniMax M2.780.9%estimated ± 5.6 pp, low confidence
75Gemini 3.5 Flash-Lite80.5%estimated ± 5.6 pp, low confidence
76Grok 4 Fast (Reasoning)80.5%estimated ± 5.6 pp, low confidence
77Qwen3 Max80.2%estimated ± 5.6 pp, low confidence
78GPT-5.579.6%estimated ± 5.1 pp, low confidence
79Claude Opus 4.579.1%estimated ± 5.6 pp, low confidence
80Claude Sonnet 4.5 Thinking78.6%estimated ± 5.6 pp, low confidence
81GLM-5.378.3%measured
82o3-mini78.2%estimated ± 5.6 pp, low confidence
83MiMo-V2.6-Pro78.2%estimated ± 3.9 pp, medium confidence
84Claude Haiku 5.578.2%estimated ± 3.9 pp, medium confidence
85GPT-6 Luna78.2%estimated ± 3.9 pp, medium confidence
86Gemini 3.8 Flash78.2%estimated ± 3.9 pp, medium confidence
87GLM-5.3-Flash78.2%estimated ± 3.9 pp, medium confidence
88Composer 2.577.7%estimated ± 5.1 pp, low confidence
89Laguna M.177.6%estimated ± 5.6 pp, low confidence
90Laguna XS.277.6%estimated ± 5.6 pp, low confidence
91GLM-4.577.5%estimated ± 5.6 pp, low confidence
92Grok Code Fast 176.9%estimated ± 5.6 pp, low confidence
93Claude 3.7 Sonnet Thinking76.8%estimated ± 5.6 pp, low confidence
94GLM-5.276.7%estimated ± 5.1 pp, low confidence
95GPT-4.1 mini76.7%estimated ± 5.6 pp, low confidence
96Claude 3.7 Sonnet76.6%estimated ± 5.6 pp, low confidence
97Mistral Large 376.5%estimated ± 5.6 pp, low confidence
98GPT-4.176.5%estimated ± 5.6 pp, low confidence
99o176.4%estimated ± 5.6 pp, low confidence
100Claude 3.5 Sonnet76.3%estimated ± 5.6 pp, low confidence
101Gemini 2.0 Flash76.2%estimated ± 5.6 pp, low confidence
102GPT-4o76.2%estimated ± 5.6 pp, low confidence
103GPT-4.1 nano76.2%estimated ± 5.6 pp, low confidence
104Claude 3.5 Haiku76.2%estimated ± 5.6 pp, low confidence
105Gemini 1.5 Pro76.2%estimated ± 5.6 pp, low confidence
106Claude Haiku 4.576.2%measured
107Gemini 1.5 Flash76.2%estimated ± 5.6 pp, low confidence
108GPT-4o mini76.1%estimated ± 5.6 pp, low confidence
109Gemini 3.6 Flash75.4%estimated ± 5.1 pp, low confidence
110Kimi K373.7%measured

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General