CharXiv Reasoning (no tools)
Can the model extract and reason over information in images or documents?
Vision15 ranked models22 reported values4 reportsHigher is better
Top rankings
One row per model, using its best reported score across effort settings.
| Rank | Model | Best score | Best reported setting |
|---|---|---|---|
| 1 | Gemini 3.8 FlashGoogle | 86.2% | Reported configurationGoogle DeepMind report1 value · 1 reportSep 2, 2026 · source |
| 2 | GPT-5.6 TerraOpenAI | 85.9% | Reported configurationGoogle DeepMind report2 values · 2 reportsSep 2, 2026 · source |
| 3 | GPT-5.6 SolOpenAI | 85.8% | Reported configurationGoogle DeepMind report1 value · 1 reportSep 2, 2026 · source |
| 4 | Gemini 3.6 FlashGoogle | 85.2% | Reported configurationGoogle report2 values · 2 reportsAug 13, 2026 · source |
| 5 | Gemini 3.7 FlashGoogle | 84.5% | Reported configurationGoogle DeepMind report2 values · 2 reportsSep 2, 2026 · source |
| 6 | Gemini 3.5 FlashGoogle | 84.2% | Reported configurationGoogle report2 values · 2 reportsJul 21, 2026 · source |
| 7 | GPT-5.5OpenAI | 84.1% | Reported configurationGoogle report1 value · 1 reportMay 19, 2026 · source |
| 8 | Claude Opus 5Anthropic | 83.7% | Reported configurationGoogle DeepMind report1 value · 1 reportSep 2, 2026 · source |
| 9 | Gemini 3.1 ProGoogle | 83.3% | Reported configurationGoogle report2 values · 2 reportsJul 21, 2026 · source |
| 10 | GPT-5.6 LunaOpenAI | 82.7% | Reported configurationGoogle report1 value · 1 reportJul 21, 2026 · source |
| 11 | Claude Opus 4.7Anthropic | 82.1% | Reported configurationGoogle report1 value · 1 reportMay 19, 2026 · source |
| 12 | Grok 4.5SpaceXAI | 81.6% | Reported configurationGoogle report1 value · 1 reportJul 21, 2026 · source |
Effort curve
Every sourced cost-linked effort value for this exact version. Lines connect complete sweeps only.
No cost-linked effort sweep for this version.
Definition and comparison boundarypublic methodology
Can the model extract and reason over information in images or documents?
Visual perception and multimodal reasoning under the reported protocol. Higher is better. The value is the percentage reported in this lab's table.
Harness, tools, prompts, attempt count, and benchmark version can materially change the result. Compare within one reporting context.
No tools. No tools. Gemini and GPT-5.6 Terra cells are Google-computed. Information synthesis from complex charts; no tools.
Method / source