benchgap
Vision & documents

MathVision w/ Python leaderboard

As of 2026-11-21, the highest measured score on MathVision w/ Python is 97.8% by Kimi K3. 22 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Kimi K397.8%measured
2Qwen3.8 Max97.7%measured
3Claude Mythos 597.7%estimated ± 1.0 pp, low confidence
4Kimi K3.1 Pro97.6%estimated ± 1.0 pp, low confidence
5Glimpse97.3%estimated ± 1.0 pp, low confidence
6Gemini 3.8 Flash97.0%estimated ± 1.3 pp, low confidence
7Inkling 2 Small96.6%estimated ± 0.5 pp, medium confidence
8Seed 2.1 Pro96.5%estimated ± 0.5 pp, medium confidence
9LongCat-2.596.5%measured
10Claude Opus 4.796.3%estimated ± 1.0 pp, low confidence
11Qwen3.8-Omni-Flash96.2%measured
12Claude Opus 4.895.7%estimated ± 1.0 pp, low confidence
13Qwen3.8-Flash-Next95.7%measured
14GLM-5.3-Flash95.4%estimated ± 1.0 pp, low confidence
15Qwen3.7 Plus95.3%estimated ± 0.5 pp, medium confidence
16Seed 2.1 Turbo95.2%estimated ± 0.5 pp, medium confidence
17Gemini 3.7 Flash95.0%estimated ± 1.0 pp, low confidence
18Muse Spark 1.194.9%estimated ± 1.0 pp, low confidence
19Claude Sonnet 594.8%estimated ± 1.0 pp, low confidence
20Qwen3.8-27B94.6%measured
21Inkling 2 Mini94.6%estimated ± 0.5 pp, low confidence
22Qwen3.5 397B94.3%estimated ± 0.5 pp, low confidence
23Qwen3.6 Plus94.0%estimated ± 0.5 pp, low confidence
24dots3-note Preview93.8%estimated ± 0.5 pp, low confidence
25Kimi K2.693.6%estimated ± 0.5 pp, low confidence
26Gemini 3 Pro93.2%estimated ± 0.5 pp, low confidence
27Qwen3.5-122B-A10B93.0%estimated ± 0.5 pp, low confidence
28Qwen3.5-27B92.8%estimated ± 0.5 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General