CFE-Bench (Multimodal): leaderboard
Metric: Question accuracy (%): share of the 144 multimodal CFE-Bench problems (university STEM problems with diagrams, plots or schematics, kept only when the image is needed) for which every annotated answer variable is correct; answers extracted and verified per annotated target variable by a GPT-5-mini judge (gpt-5-mini-2025-08-07) against expert-annotated values; chain-of-thought prompting, thinking capped at 16,000 tokens; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 48.61 | #54 |
| 2 | Gemini 3 Pro (Preview) | 48.61 | #64 |
| 3 | Gemini 3 Flash (Preview) | 48.61 | #78 |
| 4 | Qwen 3.5 397B A17B | 45.14 | #141 |
| 5 | Qwen 3.5 Plus | 44.44 | #123 |
| 6 | GPT-5.2 | 43.75 | #105 |
| 7 | Claude Opus 4.6 | 36.81 | #60 |
| 8 | Claude Opus 4.5 | 30.56 | #79 |
| 9 | Grok 4 | 29.17 | #169 |
| 10 | Grok 4.1 Fast (Reasoning) | 26.39 | #208 (Grok 4.1 Fast) |
| 11 | Claude Sonnet 4.5 | 18.75 | #138 |
| 12 | Qwen 3 VL 32B Instruct | 10.42 | #276 |
| 13 | Llama 4 Maverick | 9.72 | #451 |
| 14 | GLM-4.6V | 7.64 | #309 |
| 15 | Gemma 3 27B (IT) | 2.78 | #509 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=cfe-bench-multimodal · How It Works · Data refreshed daily, snapshot 2026-10-11.