LLM Stats (AI2D): leaderboard
Multiple-choice science questions with diagrams (food webs, geological cross-sections, biological processes), measuring interpretation of scientific illustrations.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Estimated already saturated. 35 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet | 94.7 | #350 |
| 2 | Qwen 3.6 Plus | 94.4 | #120 |
| 3 | GPT-4o | 94.2 | #343 |
| 4 | Qwen 3.5 122B A10B | 93.3 | #173 |
| 5 | Mistral Small 3.2 | 92.91 | #605 |
| 6 | Qwen 3.5 27B | 92.9 | #187 |
| 7 | Qwen 3.6 35B A3B | 92.7 | #226 |
| 8 | Qwen 3.5 35B A3B | 92.6 | #256 |
| 9 | Llama 3.2 90B Instruct | 92.3 | #601 |
| 10 | Llama 3.2 11B Instruct | 91.1 | #1126 |
| 11 | Qwen 3 VL 235B A22B Instruct | 89.7 | #273 |
| 12 | Qwen 3 VL 32B Instruct | 89.5 | #283 |
| 13 | Qwen 3 VL 235B A22B (Thinking) | 89.2 | #238 (Qwen 3 VL 235B A22B) |
| 14 | Seed 1.8 | 89.1 | #139 |
| 15 | EXAONE 4.5 33B | 89 | #396 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-ai2d · How It Works · Data refreshed daily, snapshot 2026-10-20.