HugAgent - ATI: leaderboard
HugAgent - ATI. HugAgent EMNLP2026: participant-specific interview context, belief-state/update/direction accuracy and ATI. Released test items from54participants; supplied person context, not the model answering about itself. Six named standalone baselines only; Human ceiling, full-context RAG, Generative Agents and global-majority controls retained in source grid, not model scores. Exact decoding/score scripts linked in native README; no new checkpoint inferred. Metric: ATI index (0-100). Higher is better.
Metric: ATI index (0-100). Source: jajamoa.github.io. Saturation forecast: Not forecast. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.3 70B | 69.84 | #1498 |
| 2 | Claude Sonnet 4.5 | 67.39 | #250 |
| 3 | GPT-4o | 67.29 | #741 |
| 4 | DeepSeek R1 | 67.2 | #473 |
| 5 | Gemini 2.5 Pro | 64.29 | #266 |
| 6 | GPT-5 Mini | 61.53 | #348 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=hugagent-ati · How It Works · Data refreshed daily, snapshot 2026-11-13.