SPUR - Qualitative Reasoning: leaderboard
Metric: Accuracy (%) on SPUR qualitative reasoning multiple-choice questions about multi-panel biomedical experimental figures from PubMed Central papers (questions drafted by GPT-4o, kept only when GPT-4o failed them in at least six of ten text-only attempts, then expert-reviewed): the questions ask the model to combine visual evidence, domain knowledge and experimental design to interpret biological significance; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 90.31 |
| 2 | Claude 3.7 Sonnet (Thinking) | 87.58 |
| 3 | Gemini 2.5 Pro (Preview 06-05) | 86.54 |
| 4 | GPT-5.1 | 86.52 |
| 5 | Llama 4 Maverick | 84.64 |
| 6 | O4 Mini (High) | 84.33 |
| 7 | GLM-4.5V | 80.94 |
| 8 | Seed-1.6 | 80.31 |
| 9 | InternVL3-78B | 75.24 |
| 10 | Qwen 3 VL 30B A3B (Thinking) | 75.16 |
| 11 | Grok 4.1 Fast | 73.44 |
| 12 | Qwen 2.5 VL 72B Instruct | 73.1 |
| 13 | Mistral Small 3.1 | 72.82 |
| 14 | Ministral-3-14B-Instruct-2512 | 72.5 |
| 15 | Ministral-3-8B-Instruct-2512 | 70.85 |
Interactive version: theaggregate.ai/benchmark?slug=spur-qualitative-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-07.