NYT Connections Original — leaderboard

Original version: 436 standard NYT Connections puzzles without modifications. Models get one attempt (humans get four). ~71% average human solve rate.

Metric: Score (%). Source: github.com. Status: saturation imminent. 29 models tracked.

Top models

#ModelScore
1Human Expert100
2O190.7
3O1 Preview87.1
4O3 Mini72.4
5Median Human71
6DeepSeek R154.4
7O1 Mini42.2
8Gemini 2.0 Flash (01-21) (Thinking)37
9GPT-4 Turbo28.3
10GPT-4o (2024-11-20)27.9
11GPT-4o (2024-08-06)26.5
12Llama 3.1 405B26.3
13Claude 3.5 Sonnet (20241022)25.9
14Claude 3 Opus24.8
15Llama 3.3 70B23.7

Interactive version: theaggregate.ai/benchmark?slug=nyt-connections-original · How the rankings work · Data refreshed daily, snapshot 2026-07-22.