Beyond Rating: leaderboard
Metric: Weakness Max-Recall (from 0-1, times 100): share of a single human reviewer's weakness points that the generated review's weakness points cover, taking the best-covered reviewer per paper, on 1,000 test papers from ICLR 2024-2026 and NeurIPS 2022-2025 with three to five high-confidence human reviews each, the model writing a full review from the parsed paper; review points are split into atomic claims and matched to the human reviews' claims by Qwen3-235B; higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.5 | 46 |
| 2 | GPT-5.2 | 42 |
| 3 | DeepSeek V3.2 | 36 |
| 4 | Qwen 3 235B A22B 2507 Instruct | 32 |
| 5 | Qwen 3 30B A3B 2507 Instruct | 29 |
| 6 | Gemini 3 Pro (Preview) | 28 |
| 7 | Qwen 3 8B | 25 |
| 8 | Llama 3.1 8B Instruct | 16 |
| 9 | Llama 3.1 70B Instruct | 16 |
Interactive version: theaggregate.ai/benchmark?slug=beyond-rating · How It Works · Data refreshed daily, snapshot 2026-10-07.