TraceBench (High Noise, Program): leaderboard
TraceBench (2026) generates controlled time-series root-cause attribution tasks from interpretable mechanical simulators, where an LLM agent decides whether and which system parameter changed; chance is about 19 percent. Episode accuracy under high observation noise when the agent submits a Python program, in percent; higher is better.
Metric: Accuracy (%; high observation noise; the agent must submit a Python program mapping each observation to a predicted label; controlled root-cause attribution tasks generated from three interpretable mechanical simulators (BallDrop, BounceBall, MassSlide): the agent reads time-series observations and decides whether a system parameter was altered and which one; chance 0.189, mean over three simulators). Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (High) | 73.3 | #48 (GPT-5.5) |
| 2 | Claude Opus 4.6 (High) | 59.3 | #100 (Claude Opus 4.6) |
| 3 | Gemini 3.1 Pro (Preview) (High) | 54.7 | #91 (Gemini 3.1 Pro (Preview)) |
| 4 | MiniMax-M2.7 | 35.3 | #519 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=tracebench-high-noise-program · How It Works · Data refreshed daily, snapshot 2026-11-12.