CLIR-Bench - Cross-Variable Reasoning: leaderboard
Metric: Accuracy (%) on asynchronous cross-variable reasoning, 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 2031. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 Mini | 63.33 |
| 2 | Gemini 2.5 Flash | 43.33 |
| 3 | Qwen 3.5 9B | 27.5 |
| 4 | Qwen 3.5 4B | 24.33 |
| 5 | Qwen 3.6 27B | 21.67 |
| 6 | DeepSeek V4 Flash | 18.33 |
| 7 | Gemma 3 27B (IT) | 16 |
| 8 | Kimi K2.6 | 15 |
| 9 | Gemma 4 12B (IT) | 6.17 |
Interactive version: theaggregate.ai/benchmark?slug=clir-bench-cross-variable-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-29.