LexKairos (Thinking Mode) - Bounded Event Retrieval: leaderboard
Metric: Accuracy (%; 594 items: retrieve the case events inside a time window, single- and multi-label; thinking mode, built-in reasoning enabled at high effort with the vanilla prompt; temperature 0, 8,000 output tokens). Source: arxiv.org. Saturation forecast: Around December 2026. 5 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash (High) | 88.72 |
| 2 | DeepSeek V4 Flash (High) | 86.87 |
| 3 | GPT-5.4 (High) | 86.2 |
| 4 | Qwen 3.5 9B (Thinking) | 69.19 |
Interactive version: theaggregate.ai/benchmark?slug=lexkairos-thinking-mode-bounded-event-retrieval · How It Works · Data refreshed daily, snapshot 2026-09-29.