AJ-Bench (LLM-as-a-Judge) - PPT: leaderboard
Metric: F1 (%) on PowerPoint GUI tasks (OSWorld; 21 tasks, 42 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) | 76.1 |
| 2 | Claude Sonnet 4.5 | 75.61 |
| 3 | Gemini 2.5 Pro | 68.72 |
| 4 | Kimi K2 0905 | 65.53 |
| 5 | Grok 4 | 61.11 |
| 6 | GLM-4.6 | 60.82 |
| 7 | Claude Opus 4.5 | 59.21 |
| 8 | DeepSeek V3.2 | 58.38 |
| 9 | GPT-5 | 51.9 |
| 10 | Qwen 3 235B A22B | 45.5 |
| 11 | LongCat-Flash-Chat | 45.33 |
| 12 | GPT-5 Mini (Low) | 45.05 |
| 13 | GPT-5.1 | 41.9 |
Interactive version: theaggregate.ai/benchmark?slug=aj-bench-llm-as-a-judge-ppt · How It Works · Data refreshed daily, snapshot 2026-10-07.