Sci-Rho - Worst-Case: leaderboard
Sci-Rho (Science Rhobustness, 2026) is a dynamic benchmark of image-grounded STEM problems in five subjects and seven languages, where each expert-written template generates ten visually and numerically varied instances. This board is worst-case accuracy (%), the share of templates answered correctly on every generated variant, averaged over the seven languages; higher is better.
Metric: Worst-case accuracy (%) over all seven languages on 606 expert-written executable templates per language (mathematics, physics, chemistry, biology and computer science, many drawn from olympiad problems), each rendered as 10 visually and numerically varied image-grounded instances: the share of templates a model answers correctly on every one of its 10 generated variants, averaged over languages; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (Non-reasoning) | 73.9 | #123 (GPT-5.4) |
| 2 | Gemini 2.5 Pro | 62.5 | #266 |
| 3 | Gemini 2.5 Flash (Non-reasoning) | 56.6 | #453 (Gemini 2.5 Flash) |
| 4 | Qwen 3.5 27B (Non-reasoning) | 51.2 | #359 (Qwen 3.5 27B) |
| 5 | Qwen 3.5 122B A10B (Non-reasoning) | 43.6 | #331 (Qwen 3.5 122B A10B) |
| 6 | Llama 4 Scout Instruct | 33.7 | #1543 |
| 7 | Qwen 3 VL 8B Instruct | 29.5 | #981 |
| 8 | Gemma 3 12B (IT) | 18.7 | #1837 |
| 9 | Molmo2-8B | 15.5 | #1483 |
| 10 | Gemma 3 4B (IT) | 7.2 | #3831 |
| 11 | Qwen 3 VL 4B Instruct | 3.3 | #1279 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=sci-rho-worst-case · How It Works · Data refreshed daily, snapshot 2026-11-13.