SkillTV-Bench (Trajectory Verification): leaderboard
SkillTV-Bench (Shanghai Jiao Tong University, 2026) evaluates how well a model judges whether a skill-augmented agent trajectory succeeded, over 203 held-out real agent executions with their task-time skills and inspectable artifacts. Balanced accuracy is the mean of the success-class and failure-class accuracy of the pass or fail verdict, in percent, higher is better.
Metric: Balanced accuracy (%; mean of success-class and failure-class accuracy for the judge's pass/fail verdict over 203 held-out agent trajectories in nine domains). Source: arxiv.org. Saturation forecast: Around February 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 58.6 | #91 |
| 2 | Claude Sonnet 4.6 | 56 | #158 |
| 3 | GPT-5.2 | 47.8 | #172 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=skilltv-bench-trajectory-verification · How It Works · Data refreshed daily, snapshot 2026-11-12.