STEMVerse - Physics - Bloom Level: Evaluate: leaderboard
STEMVerse physics problems labelled with the Bloom cognitive level 'Evaluate' (judging or justifying claims), drawn from existing STEM benchmarks pooled into one set; labels come from GPT-4o and were checked by an expert audit. Few-shot prompting in the MegaScience style; accuracy in percent, higher is better. Li et al. (ZGCA and NTU), 2026 preprint.
Metric: Few-shot accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3-14B-Base | 50 | #1608 |
| 2 | Qwen3-8B-Base | 43.75 | #1728 |
| 3 | Qwen 2.5 14B Instruct | 37.5 | #1745 |
| 4 | Llama 3.1 8B Instruct | 31.25 | #4277 |
| 5 | Qwen 2.5 7B Instruct | 31.25 | #2583 |
| 6 | Llama 3.2 3B Instruct | 31.25 | #7185 |
| 7 | Qwen 2.5 7B | 31.25 | #2718 |
| 8 | Qwen 2.5 14B | 31.25 | #1794 |
| 9 | Qwen 2.5 3B Instruct | 25 | #5206 |
| 10 | Llama 3.2 3B | 25 | #7877 |
| 11 | Llama 3.1 8B | 18.75 | #4900 |
| 12 | Qwen3-4B-Base | 18.75 | #2444 |
| 13 | Qwen 2.5 3B | 12.5 | #5214 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=stemverse-physics-bloom-level-evaluate · How It Works · Data refreshed daily, snapshot 2026-11-12.