ReLE - Overall: leaderboard

Metric: Accuracy (%). Source: nonelinear.com. 162 models tracked.

Top models

#ModelScore
1Qwen 3.7 Max76.9
2Claude Opus 576.8
3Gemini 3.8 Flash76.2
4Kimi K375.6
5Qwen 3.8 Max75.4
6Gemini 3.7 Flash75.4
7Qwen 3.6 Max Preview75.4
8GPT-5.575.3
9Gemini 3.1 Pro (Preview)75.2
10GLM-5.375.2
11DeepSeek V4 Pro74.9
12Claude Opus 4.8 (Thinking)74.7
13DeepSeek V4.1 Flash74.4
14GPT-5.6 Pro Sol74.4
15GLM-5.3 Flash74.2

Interactive version: theaggregate.ai/benchmark?slug=rele-overall · How It Works · Data refreshed daily, snapshot 2026-09-19.