LongJudgeBench (Reference) - RealDeepResearch: leaderboard
Metric: Judge accuracy (%) on RealDeepResearch (640 expert-annotated Chinese and English deep-research reports in DOCX form, pointwise): agreement of the LLM judge's pairwise preferences (from pointwise scores, pairwise choices or listwise ranks) or correctness labels with the human reference judgments, averaged within query groups, temperature 0, judging with reference answers or expert reports; higher is better. Source: arxiv.org. Saturation forecast: Around August 2027. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GLM-5.1 | 50.92 |
| 2 | GPT-5.2 | 50.85 |
| 3 | Qwen 3 Max | 48.08 |
| 4 | Kimi K2.6 | 45.77 |
| 5 | Qwen 3 32B (Thinking) | 43.73 |
| 6 | DeepSeek V4 Flash | 40.67 |
| 7 | Qwen 3 32B (Non-reasoning) | 32.71 |
| 8 | GPT-4o Mini | 24.35 |
Interactive version: theaggregate.ai/benchmark?slug=longjudgebench-reference-realdeepresearch · How It Works · Data refreshed daily, snapshot 2026-09-29.