CLIR-Bench - Temporal Grounding: leaderboard

Metric: Accuracy (%) on temporal grounding (locate the time points or intervals that answer the question), 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 October 2027. 15 models tracked.

Top models

#ModelScore
1GPT-5.4 Mini58.33
2Kimi K2.650
3Gemma 4 12B (IT)45
4Gemini 2.5 Flash41.67
5Qwen 3.5 4B29.17
6DeepSeek V4 Flash28.33
7Qwen 3.5 9B27.17
8Qwen 3.6 27B23.17
9Gemma 3 27B (IT)22.33

Interactive version: theaggregate.ai/benchmark?slug=clir-bench-temporal-grounding · How It Works · Data refreshed daily, snapshot 2026-09-29.