Benchmarks / Reasoning

Roboflow Visual Reasoning

Roboflow Vision Evals: image questions requiring arithmetic, comparisons, spatial relations and multi-step inference (the answer is never written in the image). LLM-judge accuracy, mean of runs.

57models
–Open Index weight
–last updated
Roboflowpublisher

Roboflow Visual Reasoning

Higher is better · best effort per model

Chart settings
57 of 57 selected

full table below

#ModelEffortScoreCost / taskHarness
1Gemini 4high91.4%$0.013–
2GPT-6 Astrahigh91.2%$0.021–
3GPT-6.1 Solhigh88.7%$0.0042–
4Opus 5.5high85.9%$0.011–
5Muse Wonderlow85.2%$0.024–
6Gemini 3.8 Flashhigh84.5%$0.023–
7Sonnet 5.5high83.9%$0.0061–
8Gemini 3.5 Flashlow82.1%$0.0082–
9Gemini 3.7 Flashhigh81.9%$0.0050–
10Gemini 3.6 Flashhigh81.0%$0.0091–
11Qwen 3.8 Maxhigh80.3%$0.011–
12GPT-6 Solhigh77.7%$0.0061–
13Muse Spark 1.1high76.6%$0.013–
14Muse Spark 1.2high75.7%$0.012–
15Haiku 5.5high74.8%$0.0006–
16Gemini 3.1 Prohigh74.8%$0.021–
17Gemini 3 Flashhigh74.2%$0.0040–
18Opus 5high74.2%$0.018–
19Kimi K3high74.2%$0.026–
20Muse Spark 1.3low73.3%$0.0064–
21Fable 5.1high73.1%$0.028–
22GPT-5.5high72.2%$0.044–
23GPT-5.6 Solhigh71.7%$0.012–
24GPT-6 Lunahigh71.1%$0.0004–
25Qwen 3.8 Flashhigh69.5%$0.0011–
26Gemini 3.5 Flash-Litehigh68.9%$0.0042–
27Qwen 3.7 Plushigh68.2%$0.0043–
28Grok 4.7high66.9%$0.024–
29Fable 5low66.2%$0.018–
30Qwen 3.8 27Bhigh66.0%$0.0000–
31GPT-5.6 Lunahigh65.6%$0.0015–
32GPT-5.6 Terrahigh65.3%$0.0067–
33GPT-5.4 Minihigh64.0%$0.0096–
34Grok 4.6high63.8%$0.032–
35Gemini 2.5 Prohigh62.3%$0.011–
36Qwen 3.7 Flashhigh61.6%$0.0005–
37Qwen 3.5 27Bhigh61.6%$0.011–
38Grok 4.5high59.8%$0.019–
39GLM 5.3 Flashhigh59.6%$0.0002–
40Qwen 3.6 27Blow59.2%$0.0000–
41Kimi K2.6high58.9%$0.020–
42MiMo V2.6 Flashhigh58.5%$0.0008–
43Mistral Large 4high57.6%$0.013–
44MiMo V2.6 Prohigh55.9%$0.0026–
45Qwen 3.6 35B-A3Blow54.8%$0.0000–
46Qwen 3.5 35B-A3Blow54.1%$0.0000–
47Opus 4.8low53.0%$0.0078–
48Gemma 4 31Blow52.8%$0.0000–
49GLM 5 Vhigh49.7%$0.0069–
50DeepSeek V4 Flashhigh49.0%$0.0025–
51Gemma 4 26Blow47.7%$0.0000–
52Qwen 3.5 9Blow45.9%$0.0000–
53Sonnet 5low43.0%$0.0032–
54GLM 4.6 Vlow41.5%$0.0000–
55Qwen 3.5 4Blow38.6%$0.0000–
56Qwen 3high33.8%$0.0002–
57Qwen 3.5 2Blow27.1%$0.0000–