PRL-Bench: leaderboard
Metric: Score (0-100) over all tasks, research tasks built from 100 Physical Review Letters papers; a GPT-5 judge awards pre-assigned points for final answers and intermediate rubric checkpoints, summed and normalized to 0-100, mean of five runs; code interpreter allowed, search disabled; higher is better. Source: arxiv.org. Saturation forecast: Around March 2028. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 44.27 |
| 2 | Qwen 3.5 Plus | 40.05 |
| 3 | Seed 2.0 Pro | 37.83 |
| 4 | Claude Opus 4.6 | 37.4 |
| 5 | GPT-5.4 | 37.38 |
| 6 | Kimi K2.5 | 33.89 |
Interactive version: theaggregate.ai/benchmark?slug=prl-bench · How It Works · Data refreshed daily, snapshot 2026-10-07.