LexKairos - Effective Version Identification: leaderboard

Metric: Accuracy (%; 452 items: generate the statute title and version tag in force at a given date; vanilla setting, the model answers directly without extra reasoning steps (thinking disabled where the model allows it); temperature 0, 8,000 output tokens). Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.

Top models

#ModelScore
1GPT-5.4 (Non-reasoning)59.96
2Gemini 3 Flash (Minimal)59.51
3DeepSeek R147.57
4DeepSeek V4 Flash (Non-reasoning)45.58
5Qwen 3.5 9B (Non-reasoning)40.04
6Llama 3.1 8B12.83

Interactive version: theaggregate.ai/benchmark?slug=lexkairos-effective-version-identification · How It Works · Data refreshed daily, snapshot 2026-09-29.