RBench-V: leaderboard
RBench-V — visual reasoning benchmark with multi-step problems over images; overall score.
Metric: Overall Score (%). Source: evalmodels.github.io. Saturation forecast: Around May 2028. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 27.9 | #186 |
| 2 | O3 | 25.8 | #124 |
| 3 | O4 Mini | 20.9 | #180 |
| 4 | Gemini 2.5 Pro (Preview 05-06) | 20.2 | #147 |
| 5 | O1 | 16.2 | #183 |
| 6 | GPT-4.1 | 13.6 | #248 |
| 7 | GPT-4.5 | 12.6 | #274 |
| 8 | Claude 3.7 Sonnet | 11.5 | #249 |
| 9 | InternVL3-38B | 10 | #405 |
| 10 | Llama 4 Scout | 9.5 | #656 |
| 11 | MiniCPM-V-2.6 | 9.1 | #836 |
| 12 | Qwen 2.5 VL 7B | 8.3 | #643 |
| 13 | InternVL3-8B | 8.2 | #620 |
| 14 | InternVL3-14B | 8 | #505 |
| 15 | Qwen2.5-Omni-7B | 7.7 | #624 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=rbench-v · How It Works · Data refreshed daily, snapshot 2026-10-20.