Coding

VulcanBench v3 leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on VulcanBench v3 is 89.9% by Grok 4.5. 52 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Grok 4.5	89.9%	measured
2	Claude Fable 5	89.5%	measured
3	DeepSeek V4 Flash 0731	88.4%	measured
4	GPT-6 Astra	87.1%	estimated ± 0.7 pp, low confidence
5	Claude Fable 5.1	87.0%	estimated ± 0.7 pp, low confidence
6	Claude Opus 5.5	87.0%	estimated ± 0.7 pp, low confidence
7	Claude Sonnet 5.5	87.0%	estimated ± 0.7 pp, low confidence
8	Grok 4.7	87.0%	estimated ± 0.7 pp, medium confidence
9	Claude Opus 5	87.0%	measured
10	GPT-5.6 Sol	87.0%	measured
11	GPT-5.6 Terra	87.0%	measured
12	Grok 4.6	87.0%	measured
13	Muse Spark 1.2	87.0%	measured
14	Muse Spark 1.3	87.0%	estimated ± 0.7 pp, medium confidence
15	Gemini 3.8 Flash	87.0%	estimated ± 0.7 pp, medium confidence
16	Gemini 3.7 Flash	86.3%	estimated ± 5.6 pp, low confidence
17	Gemini 3.1 Pro	86.2%	estimated ± 5.6 pp, low confidence
18	GPT-5.2-Codex	85.8%	estimated ± 5.6 pp, low confidence
19	GPT-5.6 Luna	85.5%	measured
20	DeepSeek V4 Pro 0813	85.4%	estimated ± 5.6 pp, low confidence
21	GPT-5.3 Codex	85.2%	estimated ± 5.6 pp, low confidence
22	Qwen3.7 Max	85.1%	estimated ± 5.6 pp, low confidence
23	Kimi K2.6	84.9%	estimated ± 5.6 pp, low confidence
24	Nemotron 3 Ultra	84.3%	estimated ± 5.6 pp, low confidence
25	Qwen3.6 Plus	84.3%	estimated ± 5.6 pp, low confidence
26	Inkling-Small	84.2%	estimated ± 5.6 pp, low confidence
27	Muse Spark 1.1	84.2%	estimated ± 5.6 pp, low confidence
28	Gemini 3 Flash	84.0%	estimated ± 5.6 pp, low confidence
29	GPT-5.1-Codex	84.0%	estimated ± 5.6 pp, low confidence
30	Inkling	83.9%	estimated ± 5.6 pp, low confidence
31	Claude Opus 4.7	83.7%	estimated ± 5.6 pp, low confidence
32	Grok 4.3	83.3%	estimated ± 5.6 pp, low confidence
33	Grok 4.20	83.2%	estimated ± 5.6 pp, low confidence
34	GPT-5.4 nano	83.0%	estimated ± 5.6 pp, low confidence
35	Ling 3.0 Flash	83.0%	estimated ± 5.6 pp, low confidence
36	GPT-5.1-Codex-Max	82.7%	estimated ± 5.6 pp, low confidence
37	Claude Opus 4.8	82.7%	estimated ± 5.1 pp, low confidence
38	Qwen3.8-27B	82.6%	measured
39	Qwen3.5 Flash	82.6%	estimated ± 5.6 pp, low confidence
40	GLM-4.7	82.0%	estimated ± 5.6 pp, low confidence
41	MiniMax M3	82.0%	estimated ± 5.6 pp, low confidence
42	Claude Sonnet 4.6	81.9%	estimated ± 5.6 pp, low confidence
43	GPT-5.4 mini	81.6%	estimated ± 5.6 pp, low confidence
44	MiMo-V2.5	81.6%	estimated ± 5.6 pp, low confidence
45	GLM-5.1	81.6%	estimated ± 5.6 pp, low confidence
46	MiMo-V2.5-Pro	81.6%	estimated ± 5.6 pp, low confidence
47	GLM-4.6	81.4%	estimated ± 5.6 pp, low confidence
48	Qwen3.8 Max	81.2%	measured
49	GLM-5.3-Flash	81.1%	estimated ± 5.6 pp, low confidence
50	Gemini 3.1 Flash-Lite	80.9%	estimated ± 5.6 pp, low confidence
51	MiniMax M2.7	80.9%	estimated ± 5.6 pp, low confidence
52	Gemini 3.5 Flash-Lite	80.5%	estimated ± 5.6 pp, low confidence
53	GPT-5.5	79.6%	estimated ± 5.1 pp, low confidence
54	GLM-5.3	78.3%	measured
55	Laguna M.1	77.6%	estimated ± 5.6 pp, low confidence
56	Laguna XS.2	77.6%	estimated ± 5.6 pp, low confidence
57	GLM-4.5	77.5%	estimated ± 5.6 pp, low confidence
58	Grok Code Fast 1	76.9%	estimated ± 5.6 pp, low confidence
59	GLM-5.2	76.7%	estimated ± 5.1 pp, low confidence
60	Claude Haiku 4.5	76.2%	measured
61	Gemini 3.6 Flash	75.4%	estimated ± 5.1 pp, low confidence
62	Claude Sonnet 5	75.2%	estimated ± 0.7 pp, low confidence
63	Kimi K3	73.7%	measured
64	Kimi K2.7 Code	72.0%	estimated ± 5.1 pp, low confidence
65	Gemini 3.5 Flash	71.2%	estimated ± 5.1 pp, low confidence
66	Composer 2.5	70.0%	estimated ± 0.7 pp, low confidence

VulcanBench v3 leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General