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

Verifying agent trajectories on AJ-Bench (2026), the Excel subdomain (38 trajectories of OSWorld spreadsheet 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 Excel GUI tasks (OSWorld; 19 tasks, 38 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
1Claude Opus 4.575.77#160
2Grok 475.52#319
3Gemini 3 Pro (Preview)74.28#119
4GLM-4.672.49#498
5Claude Sonnet 4.571.24#261
6DeepSeek V3.270.12#419
7Gemini 2.5 Pro66.67#276
8GPT-5 MiniLow64.36#364
9Kimi K2 090563.9#570
10GPT-561.78#178
11GPT-5.160.33#240
12LongCat-Flash-Chat55.88#818
13Qwen 3 235B A22B53.97#628

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-excel · How It Works · Data refreshed daily, snapshot 2026-11-21.