AJ-Bench (LLM-as-a-Judge) - FileSystem: leaderboard

Verifying agent trajectories on AJ-Bench (2026), the FileSystem subdomain (129 trajectories of MCPMark file-manipulation tasks): the judge decides whether each trajectory succeeded and is scored against binary human-verified labels. In this setting the judge reads the trajectory only, with no tools or environment access. F1 in percent; higher is better.

Metric: F1 (%) on FileSystem data-system tasks (MCPMark; 24 tasks, 129 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 February 2027. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Grok 475.7#319
2Gemini 3 Pro (Preview)75.69#119
3Claude Sonnet 4.569.26#261
4Gemini 2.5 Pro66.1#276
5Claude Opus 4.566.06#160
6Qwen 3 235B A22B64.66#628
7GLM-4.660.86#498
8GPT-5 MiniLow60.41#364
9DeepSeek V3.260.31#419
10LongCat-Flash-Chat59.13#818
11GPT-559.09#178
12Kimi K2 090555.96#570
13GPT-5.146.27#240

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

Interactive version: theaggregate.ai/benchmark?slug=aj-bench-llm-as-a-judge-filesystem · How It Works · Data refreshed daily, snapshot 2026-11-21.