Beyond Rating - Weakness Focus Divergence: leaderboard
Beyond Rating (2026) evaluates AI peer reviewers on the text of their reviews: generated reviews of ICLR and NeurIPS papers are split into atomic points and matched by Qwen3-235B against filtered high-confidence human reviews. This board is the KL divergence between the human and generated distributions of weakness points over eight review dimensions, with no upper limit; lower is better.
Metric: KL divergence of the distribution of the generated reviews' weakness points over eight review dimensions (novelty, soundness, experiments, clarity and others) from that of the human reviews' weakness points, over the test set, 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; lower is better. Source: arxiv.org. Saturation forecast: Around May 2027. 15 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 0.05 | #261 |
| 2 | SEA-E-7B | 0.05 | |
| 3 | DeepReviewer-14B | 0.07 | #844 |
| 4 | GPT-5.2 | 0.09 | #185 |
| 5 | CycleReviewer-70B | 0.09 | #1880 |
| 6 | AI Scientist + GPT-5 | 0.09 | |
| 7 | AgentReview + GPT-5 | 0.12 | |
| 8 | DeepSeek V3.2 | 0.13 | #419 |
| 9 | Qwen 3 8B | 0.15 | #1873 |
| 10 | OpenReviewer-8B | 0.16 | |
| 11 | Gemini 3 Pro (Preview) | 0.18 | #119 |
| 12 | Qwen 3 235B A22B 2507 Instruct | 0.22 | #616 |
| 13 | Llama 3.1 8B Instruct | 0.3 | #4600 |
| 14 | Qwen 3 30B A3B 2507 Instruct | 0.3 | #1100 |
| 15 | Llama 3.1 70B Instruct | 0.38 | #1571 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=beyond-rating-weakness-focus-divergence · How It Works · Data refreshed daily, snapshot 2026-11-21.