JFTA-Bench - Level 1: leaderboard
Metric: Correct rate (%) on the Level 1 entries (one or two underlying failures) of the 372-entry JFTA-Bench evaluation subset (sampled per level from all fault trees): the model, acting as a ReAct assistant, reads a fault tree in the JFTA text format and queries a trained user simulator that answers vaguely and, in rollback scenarios, revises an earlier observation; an entry succeeds only when the confirmed set of fixed faults equals the ground-truth set after rollback; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 70.49 | #145 |
| 2 | Claude Sonnet 4.5 | 65.57 | #138 |
| 3 | GPT-5 | 56.56 | #91 |
| 4 | DeepSeek V3.2 | 55.74 | #198 |
| 5 | Qwen 3 32B | 34.43 | #424 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=jfta-bench-level-1 · How It Works · Data refreshed daily, snapshot 2026-10-11.