benchgap
Math

MMAnswerBench leaderboard

As of 2026-11-21, the highest measured score on MMAnswerBench is 91.0% by GLM-5.2. 86 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Claude Fable 5.298.1%estimated ± 2.4 pp, low confidence
2GPT-6.1 Sol97.7%estimated ± 2.4 pp, low confidence
3GPT-6 Astra97.7%estimated ± 2.4 pp, low confidence
4GPT-6.2 Astra97.4%estimated ± 2.4 pp, low confidence
5Claude Opus 5.597.1%estimated ± 2.4 pp, low confidence
6Gemini 4 Neon97.1%estimated ± 2.4 pp, low confidence
7GPT-6.1 Terra97.0%estimated ± 2.4 pp, low confidence
8Gemini 4 Carbon96.9%estimated ± 2.4 pp, low confidence
9Claude Fable 5.196.8%estimated ± 2.4 pp, low confidence
10GPT-6 Sol96.7%estimated ± 2.4 pp, low confidence
11GPT-5.6 Sol96.5%estimated ± 2.4 pp, low confidence
12Claude Sonnet 5.596.4%estimated ± 2.4 pp, low confidence
13Gemini 4 Hydrogen96.3%estimated ± 2.4 pp, low confidence
14GPT-5.5 Pro96.2%estimated ± 2.4 pp, low confidence
15Claude Fable 596.0%estimated ± 2.4 pp, low confidence
16Gemini 4 Argon95.7%estimated ± 2.4 pp, low confidence
17Claude Opus 595.6%estimated ± 2.4 pp, low confidence
18GPT-5.595.6%estimated ± 2.4 pp, low confidence
19GPT-5.6 Terra95.5%estimated ± 2.4 pp, low confidence
20GPT-5.4 Pro94.8%estimated ± 2.4 pp, low confidence
21Claude Opus 4.894.2%estimated ± 2.4 pp, low confidence
22GPT-6 Luna93.9%estimated ± 2.4 pp, low confidence
23GPT-5.6 Luna93.8%estimated ± 2.4 pp, low confidence
24GPT-5.493.8%estimated ± 2.4 pp, low confidence
25Claude Haiku 5.593.0%estimated ± 2.4 pp, low confidence
26Qwen3.8 Max92.9%estimated ± 2.4 pp, low confidence
27Muse Spark 1.392.7%estimated ± 2.4 pp, low confidence
28Gemini 3.7 Flash92.1%estimated ± 2.4 pp, low confidence
29Claude Opus 4.791.7%estimated ± 2.4 pp, low confidence
30GLM-5.391.3%estimated ± 2.4 pp, low confidence
31Gemini 3.8 Flash91.3%estimated ± 2.4 pp, low confidence
32GLM-5.291.0%measured
33GPT-5.291.0%estimated ± 2.4 pp, low confidence
34DeepSeek V4.1 Flash 101791.0%estimated ± 2.4 pp, low confidence
35Claude Opus 4.690.6%estimated ± 2.4 pp, low confidence
36Grok 4.690.6%estimated ± 2.4 pp, low confidence
37Claude Sonnet 590.5%estimated ± 2.4 pp, low confidence
38DeepSeek V4 Pro90.3%estimated ± 2.4 pp, low confidence
39Qwen3.7 Max90.3%estimated ± 2.4 pp, low confidence
40Gemini 3.5 Flash89.8%estimated ± 2.4 pp, low confidence
41Gemini 3.1 Pro89.0%estimated ± 2.4 pp, low confidence
42Gemini 3.6 Flash88.8%estimated ± 2.4 pp, medium confidence
43Grok 4.588.4%estimated ± 2.4 pp, medium confidence
44Beam 1.0 501B-A23B FP888.1%estimated ± 1.3 pp, high confidence
45Inkling 2 Small88.1%estimated ± 2.8 pp, low confidence
46GLM-5.3-Flash88.0%estimated ± 2.4 pp, medium confidence
47Kimi K2.7 Code87.6%estimated ± 2.4 pp, medium confidence
48Beam 1.0 501B-A23B87.6%estimated ± 1.3 pp, high confidence
49DeepSeek V4 Flash 073187.6%estimated ± 2.8 pp, low confidence
50Grok 4.787.3%estimated ± 2.4 pp, medium confidence
51Beam 1.0 501B-A23B NVFP487.2%estimated ± 1.3 pp, high confidence
52Gemini 3 Flash86.9%estimated ± 2.4 pp, medium confidence
53GPT-5.4 mini86.9%estimated ± 2.4 pp, medium confidence
54A.X K286.5%estimated ± 1.3 pp, high confidence
55Inkling86.5%estimated ± 1.3 pp, high confidence
56Kimi K2.686.0%measured
57GPT-5.4 nano85.2%estimated ± 2.4 pp, medium confidence
58Grok 4.384.7%estimated ± 2.4 pp, medium confidence
59Ternary Bonsai 2 27B84.2%estimated ± 1.3 pp, high confidence
60Claude Opus 4.584.0%measured
61Solar Open 284.0%estimated ± 1.3 pp, high confidence
62GLM-5.183.8%measured
63Qwen3.6 Plus83.8%measured
64Inkling-Small83.6%estimated ± 1.3 pp, high confidence
65Solar Pro 483.3%estimated ± 1.3 pp, high confidence
66o4-mini83.0%estimated ± 2.4 pp, medium confidence
67GLM-4.782.9%estimated ± 2.0 pp, low confidence
68Ripple-65B Preview82.8%estimated ± 1.3 pp, high confidence
69Qwen3.7 Plus82.6%estimated ± 2.4 pp, medium confidence
70GLM-582.5%measured
71o382.3%estimated ± 2.4 pp, medium confidence
72Muse Glimmer 30B82.3%estimated ± 1.3 pp, high confidence
73MAI-Thinking-181.9%estimated ± 1.3 pp, high confidence
74Kimi K2.581.8%measured
75Nemotron 3 Super 100B81.4%estimated ± 2.0 pp, low confidence
76Qwen3.5 397B80.9%measured
77Gemini 3.1 Flash-Lite80.9%estimated ± 2.4 pp, medium confidence
78Qwen3.6-27B80.8%measured
79Gemini 3.5 Flash-Lite80.4%estimated ± 2.4 pp, medium confidence
80Gemini 2.5 Pro80.1%estimated ± 2.4 pp, medium confidence
81Claude Sonnet 4.579.9%estimated ± 2.4 pp, medium confidence
82Ling 3.0 Flash79.7%estimated ± 1.3 pp, high confidence
83Granite 4.2 30B79.4%estimated ± 2.0 pp, low confidence
84Qwen3.6-35B-A3B78.9%measured
85Qwen3.7 Flash78.7%estimated ± 2.4 pp, low confidence
86Qwen3.5 Flash78.4%estimated ± 2.4 pp, low confidence
87K-EXAONE 2.078.2%estimated ± 1.3 pp, medium confidence
88Sarvam 105B77.7%estimated ± 2.0 pp, low confidence
89o177.5%estimated ± 2.4 pp, low confidence
90Spark-X2.6-4B76.5%estimated ± 1.3 pp, medium confidence
91GPT-4.1 mini75.5%estimated ± 2.4 pp, low confidence
92GPT-4.175.3%estimated ± 2.4 pp, low confidence
93MiniMax M2.774.2%estimated ± 1.3 pp, medium confidence
94GPT-4o73.9%estimated ± 2.4 pp, low confidence
95ZAYA1-8B73.2%estimated ± 1.3 pp, medium confidence
96MiniCPM5-2B69.3%estimated ± 1.3 pp, medium confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General