CLIR-Bench - Anchor-State Retrieval: leaderboard
The anchor-state retrieval task of CLIR-Bench (2026): accuracy in percent at reading the value of a clinical variable at a clinically defined anchor time in a serialized irregular ICU time series, 4-option multiple choice. Chance is 25; higher is better.
Metric: Accuracy (%) on anchor-state retrieval (value of one variable at a clinically defined anchor time), 4-option multiple-choice questions over serialized irregular ICU time series (Full-TS input), chance 25; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 Mini | 63.33 | #389 |
| 2 | Kimi K2.6 | 56.67 | #178 |
| 3 | Gemini 2.5 Flash | 43.33 | #453 |
| 4 | Gemma 4 12B (IT) | 39.5 | #693 |
| 5 | DeepSeek V4 Flash | 36.67 | #258 |
| 6 | Qwen 3.6 27B | 27.5 | #341 |
| 7 | Gemma 3 27B (IT) | 27.21 | #1289 |
| 8 | Qwen 3.5 4B | 26.83 | #1175 |
| 9 | Qwen 3.5 9B | 25.33 | #803 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=clir-bench-anchor-state-retrieval · How It Works · Data refreshed daily, snapshot 2026-11-13.