AJ-Bench (LLM-as-a-Judge) - Word: leaderboard
Metric: F1 (%) on Word GUI tasks (OSWorld; 12 tasks, 24 trajectories), 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) | 72.14 |
| 2 | DeepSeek V3.2 | 69.77 |
| 3 | Grok 4 | 65.26 |
| 4 | Claude Sonnet 4.5 | 61.56 |
| 5 | Gemini 2.5 Pro | 60.13 |
| 6 | Kimi K2 0905 | 55.39 |
| 7 | Claude Opus 4.5 | 51.45 |
| 8 | GLM-4.6 | 50.07 |
| 9 | GPT-5 Mini (Low) | 48.41 |
| 10 | GPT-5 | 44.81 |
| 11 | GPT-5.1 | 39.54 |
| 12 | Qwen 3 235B A22B | 36.82 |
| 13 | LongCat-Flash-Chat | 30.35 |
Interactive version: theaggregate.ai/benchmark?slug=aj-bench-llm-as-a-judge-word · How It Works · Data refreshed daily, snapshot 2026-10-07.