AJ-Bench (LLM-as-a-Judge): leaderboard
Metric: Overall F1 (%), mean of the seven subdomain F1 scores and averaged over three runs, with the judge reading the trajectory only (no tools or environment access), of AJ-Bench, judging whether agent trajectories succeeded against binary human-verified labels (155 tasks, 516 trajectories), default model configurations; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 75.05 |
| 2 | Claude Sonnet 4.5 | 69.77 |
| 3 | Grok 4 | 69.24 |
| 4 | Gemini 2.5 Pro | 68.31 |
| 5 | Claude Opus 4.5 | 66.79 |
| 6 | GLM-4.6 | 64.81 |
| 7 | DeepSeek V3.2 | 64.49 |
| 8 | Kimi K2 0905 | 64.33 |
| 9 | GPT-5 | 61.02 |
| 10 | GPT-5 Mini (Low) | 59 |
| 11 | Qwen 3 235B A22B | 58.47 |
| 12 | LongCat-Flash-Chat | 57.5 |
| 13 | GPT-5.1 | 53.5 |
Interactive version: theaggregate.ai/benchmark?slug=aj-bench-llm-as-a-judge · How It Works · Data refreshed daily, snapshot 2026-10-07.