CoCoReviewBench - Category Completeness: leaderboard

CoCoReviewBench (2026) evaluates AI peer reviewers on ICLR and NeurIPS papers against category-specific human review opinions that were filtered for correctness using reviewer, author and meta-review discussions. This board is the mean per-paper ratio of the number of categories covered by the AI review to the number covered by all human reviewers, multiplied by 100; higher is better.

Metric: Category completeness (%): the average number of review categories the AI review covers per paper as a share of the union of categories all human reviewers of the paper cover; AI reviews of 1,300 ICLR and NeurIPS papers sampled from CoCoReviewBench (100 per year, original score distribution kept), each split into atomic opinions and compared category by category with the filtered human opinions by a GPT-5-Mini judge at temperature 0, general models prompted with the AI Scientist review prompt at temperature 0.6; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 17 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-5 Mini89.97#364
2DeepReviewer-7B89.93
3GPT-5.284.49#185
4nemotron-3-nano-30B-a3BReasoning82.93#1946
5DeepReviewer-14B81.98#844
6QwQ-32B79.83#907
7Qwen 3 8BThinking75.73#1873
8Qwen 3 32BThinking73.17#972
9Qwen 2.5 7B72.4#2929
10Gemini 3 Flash68.67#175
11Gemini 3 Pro67.69#145
12Llama 3.3 70B65.49#1514
13Llama 3.1 8B62.07#5196
14SEA-E-7B51.54
15CycleReviewer-70B50.89#1880

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=cocoreviewbench-category-completeness · How It Works · Data refreshed daily, snapshot 2026-11-21.