CLIR-Bench - Threshold Forecasting: leaderboard
The threshold forecasting task of CLIR-Bench (2026): accuracy in percent at predicting whether a clinical variable will cross a clinical threshold in a future window, 4-option multiple choice over a serialized ICU time series. Chance is 25; higher is better.
Metric: Accuracy (%) on threshold forecasting (will a variable cross a clinical threshold in a future window), 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 2028. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Flash | 38.33 | #258 |
| 2 | Kimi K2.6 | 36.67 | #178 |
| 3 | GPT-5.4 Mini | 33.33 | #389 |
| 4 | Gemma 3 27B (IT) | 31.67 | #1289 |
| 5 | Gemma 4 12B (IT) | 22.17 | #693 |
| 6 | Gemini 2.5 Flash | 20 | #453 |
| 7 | Qwen 3.5 9B | 18.33 | #803 |
| 8 | Qwen 3.5 4B | 18.17 | #1175 |
| 9 | Qwen 3.6 27B | 10.5 | #341 |
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-threshold-forecasting · How It Works · Data refreshed daily, snapshot 2026-11-13.