benchgap
Vision & documents

BabyVision leaderboard

As of 2026-11-22, the highest measured score on BabyVision is 82.0% by Qwen3.8 Max. 58 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Kimi K3.1 Pro85.7%estimated ± 7.4 pp, low confidence
2Claude Mythos 585.1%estimated ± 7.4 pp, low confidence
3Qwen3.8 Max82.0%measured
4Muse Spark Open80.2%measured
5Kimi K378.3%estimated ± 3.7 pp, high confidence
6GPT-6.1 Terra76.8%estimated ± 9.1 pp, low confidence
7Muse Spark 1.176.3%measured
8Seed 2.1 Pro73.7%measured
9Gemini 3.1 Pro73.6%estimated ± 9.1 pp, low confidence
10Gemini 3.5 Flash73.0%estimated ± 9.1 pp, low confidence
11GPT-5.6 Sol71.8%estimated ± 9.1 pp, low confidence
12Gemini 3.8 Flash70.6%estimated ± 7.4 pp, low confidence
13Inkling 2 Small68.6%measured
14GPT-5.467.9%estimated ± 9.1 pp, low confidence
15GPT-5.567.9%estimated ± 9.1 pp, low confidence
16Glimpse67.8%estimated ± 11.5 pp, low confidence
17LongCat-2.567.8%estimated ± 11.5 pp, low confidence
18Sakana Fugu-Ultra67.8%estimated ± 11.5 pp, low confidence
19Sakana Fugu67.7%estimated ± 11.5 pp, low confidence
20Qwen3.8-Omni-Flash67.5%estimated ± 3.7 pp, high confidence
21GPT-5.6 Terra66.9%estimated ± 9.1 pp, low confidence
22Muse Spark66.2%estimated ± 9.1 pp, low confidence
23Qwen3.8-27B65.7%measured
24Gemini 3.7 Flash63.4%estimated ± 7.4 pp, low confidence
25Seed 2.1 Turbo62.9%measured
26Qwen3.8-Flash-Next62.8%estimated ± 3.7 pp, high confidence
27Hy462.1%estimated ± 9.1 pp, low confidence
28Qwen3.7 Plus61.7%estimated ± 3.7 pp, high confidence
29GPT-5.6 Luna61.7%estimated ± 9.1 pp, low confidence
30Grok 4.361.0%estimated ± 9.1 pp, low confidence
31MiniMax M361.0%estimated ± 9.1 pp, low confidence
32MiniMax M3.1 Flash60.8%estimated ± 11.5 pp, low confidence
33Pareto 26.960.8%estimated ± 9.1 pp, low confidence
34MiMo-V2.560.5%estimated ± 9.1 pp, low confidence
35Claude Opus 4.659.1%estimated ± 9.1 pp, low confidence
36Gemma 4 31B58.2%estimated ± 9.1 pp, low confidence
37GPT-5.4 mini57.5%estimated ± 9.1 pp, low confidence
38Step 5 Preview56.1%estimated ± 9.1 pp, low confidence
39Qwen3.5 397B55.9%estimated ± 3.7 pp, high confidence
40Hy4 flash55.6%estimated ± 9.1 pp, low confidence
41Qwen3.6-27B55.6%estimated ± 9.1 pp, low confidence
42Claude Opus 4.755.2%estimated ± 7.4 pp, low confidence
43Qwen3.6-35B-A3B54.4%estimated ± 9.1 pp, low confidence
44Grok 4.2054.2%estimated ± 9.1 pp, low confidence
45Qwen3.6 Plus53.9%estimated ± 3.7 pp, high confidence
46GLM-5.3-Flash53.4%measured
47Inkling 2 Mini53.4%measured
48Kimi K2.652.1%estimated ± 3.7 pp, high confidence
49Muse Glimmer 30B51.3%estimated ± 9.1 pp, low confidence
50Gemma 4 26B A4B50.9%estimated ± 9.1 pp, low confidence
51Claude Opus 4.850.5%estimated ± 7.4 pp, low confidence
52dots3-note Preview50.0%measured
53Gemini 3 Pro49.7%estimated ± 3.7 pp, high confidence
54Qwen3.5-122B-A10B48.5%estimated ± 3.7 pp, high confidence
55Qwen3.5-27B47.9%estimated ± 3.7 pp, high confidence
56Inkling44.6%estimated ± 7.4 pp, low confidence
57Interfaze Beta44.4%estimated ± 9.1 pp, low confidence
58Inkling-Small43.1%estimated ± 7.4 pp, low confidence
59Kimi K2.543.1%estimated ± 3.7 pp, high confidence
60Qwen3.5-35B-A3B42.3%estimated ± 3.7 pp, high confidence
61Claude Sonnet 542.3%estimated ± 7.4 pp, low confidence
62GPT-5.240.1%estimated ± 3.7 pp, high confidence
63Qwen3-VL-235B-A22B Instruct37.5%estimated ± 9.1 pp, low confidence
64Gemma 4 12B32.7%estimated ± 3.7 pp, high confidence
65Nemotron 3.5 Super VL 120B A12B29.6%measured
66Nemotron 3.5 Super VL 120B A12B NVFP429.5%measured
67Claude Sonnet 4.626.4%estimated ± 11.5 pp, low confidence
68Nemotron 3 Nano Omni 30B A3B26.4%estimated ± 11.5 pp, low confidence
69Gemini 3.1 Flash-Lite26.4%estimated ± 11.5 pp, low confidence
70Claude Opus 4.522.7%estimated ± 3.7 pp, medium confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General