Skip to content

Benchmark

CharXiv Reasoning

Vision benchmarks measure how well models understand images. Scores are percentages. Higher is better.

Models and systems measured
26
Scores recorded
47
Sources
1
IQ weight
2.13

Score distribution

82.6%
Mean
83.3%
Median
5.7%
Std dev
73.2%
10th percentile
88.9%
90th percentile
64.4%
Min
64.4%91%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.13 for chat models, measured over 25 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model or systemScore
1Claude Opus 4.7validation split · 1,000 questions · Python tools · adaptive thinking · max effort · avg@5· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.91%
2Claude Opus 4.8Fable/Mythos 5 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted · reasoning · with tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.89.9%
3Claude Opus 4.8max · with code execution · 1,000 validation questions · gpt-oss-120b high judge· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.89.9%
4Gemini 3.6 FlashGemini 3.6 Flash Model Card · July 2026 · reasoning · with tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.89.4%
5GLM 5.3 Flashwith tools· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.89.4%
6Gemini 3.7 Flashsearch + code tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.88.7%
7GPT 5.2reasoning · Python · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.88.7%
8Muse Spark 1.1Meta Model API · xhigh · with code execution · 1,000 validation questions · gpt-oss-120b high judge· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.88.4%
9Claude Sonnet 51,000 validation questions · Python + image crop · Sonnet 4.6 grader · 5 runs· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.88.3%
10Claude Sonnet 5Gemini 3.6 Flash Model Card · July 2026 · reasoning · with tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.88.3%
11Fugu UltraFugu official headline evaluation · Reasoning· Reported by Sakana AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.86.6%
12Gemini 3.8 Flashno tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.86.2%
13GPT-5.6 Solno tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.85.8%
14Claude Sonnet 4.6Python + image crop · Sonnet 5 card rerun with Sonnet 4.6 grader· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.85.3%
15Gemini 3.6 FlashGemini 3.6 Flash Model Card · July 2026 · reasoning · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.85.2%
16Gemini 3.5 FlashGemini 3.6 Flash Model Card · July 2026 · reasoning · with tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.84.9%
17GPT 5.5xhigh · with code execution · 1,000 validation questions · gpt-oss-120b high judge· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.84.8%
18Gemini 3.7 Flashno tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.84.5%
19Gemini 3.5 FlashGemini 3.5 Flash Model Card · May 2026 · reasoning · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.84.2%
20Gemini 3.5 FlashGemini 3.6 Flash Model Card · July 2026 · reasoning · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.84.2%
21GPT 5.5Gemini 3.5 Flash Model Card · May 2026 · reasoning · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.84.1%
22Claude Opus 5no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.83.7%
23Gemini 3.1 Pro PreviewGemini 3.5 Flash Model Card · May 2026 · reasoning · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.83.3%
24Gemini 3.1 Pro PreviewGemini 3.6 Flash Model Card · July 2026 · reasoning · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.83.3%
25Gemini 3.1 Pro PreviewGemini 3.6 Flash Model Card · July 2026 · reasoning · with tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.83.2%
26GPT-5.6 LunaGemini 3.6 Flash Model Card · July 2026 · reasoning · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.82.7%
27Claude Opus 4.7Gemini 3.5 Flash Model Card · May 2026 · reasoning · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.82.1%
28Claude Opus 4.7validation split · 1,000 questions · no tools · adaptive thinking · max effort · avg@5· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.82.1%
29GPT 5.2reasoning · no tools · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.82.1%
30Gemini 3.1 Pro Previewhigh · with code execution · 1,000 validation questions · gpt-oss-120b high judge· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.81.6%
31Grok 4.5Gemini 3.6 Flash Model Card · July 2026 · reasoning · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.81.6%
32Gemini 3 Pro PreviewGemini 3 Pro Model Card · results as of November 2025 · reasoning· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.81.4%
33Qwen3.5 397B A17BQwen3.5 release/model card · published setting· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.8%
34Claude Opus 4.8Fable/Mythos 5 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted · reasoning · no tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.5%
35Gemini 3 FlashGemini 3.5 Flash Model Card · May 2026 · reasoning · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.3%
36Gemini 3 Flashreasoning · no tools · 1,000 validation questions· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.3%
37Muse Glimmer 30Breasoning · high reasoning· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.78.8%
38Claude Sonnet 4.6image-cropping tool · adaptive thinking · max effort · 5 runs · original 4.6 card harness· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.77.4%
39Claude Sonnet 51,000 validation questions · no tools · Sonnet 4.6 grader · 5 runs· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.77%
40Claude Sonnet 5Gemini 3.6 Flash Model Card · July 2026 · reasoning · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.77%
41Gemini 3.1 Flash LiteGemini 3.1 Flash-Lite Model Card · high thinking · default sampling · single attempt/pass@1 · reasoning · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.73.2%
42Qwen3.5 Omni PlusQwen3.5-Omni release · published setting· Reported by Qwen, Qwen Team· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.72.5%
43Claude Sonnet 4.6Gemini 3.5 Flash Model Card · May 2026 · reasoning · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.72.4%
44Claude Sonnet 4.6no tools · adaptive thinking · max effort · 5 runs · original 4.6 card harness· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.72.4%
45Claude Sonnet 4.6no tools · Sonnet 5 card rerun with Sonnet 4.6 grader· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.71.6%
46Qwen3.5 Omni FlashQwen3.5-Omni release · published setting· Reported by Qwen, Qwen Team· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.64.4%

Not ranked

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

Model or systemScore
Muse Spark (version unspecified)Meta Model API · highest effort · version unspecified · with code execution · 1,000 validation questions · gpt-oss-120b high judge· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.88.9%

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.