AI vs Connections Logo

AI vs Connections

Comparing how different AI models perform on the NYT Connections word game. We measure accuracy, speed, and approach to understand how language models process semantic relationships.

Total Models

25

Active models being tested

Games Analyzed

703

Total games in database

Average Difficulty

3.0/5

Across all games

Top Performer

Sonar Reasoning

94% solve rate

Top Solving Models

By solve rate percentage

1.Sonar Reasoning Pro56.3%
2.Deepseek R147.8%
3.Sonar Reasoning42.2%
4.Claude 3.7 Sonnet15.5%
5.GPT-4.114.1%

Fastest Models

Average solve time in seconds

1.o31.3s
2.Mistral 7B1.5s
3.o3-mini1.5s
4.Mistral Small1.7s
5.Gemini 1.5 Flash1.7s

Longest Streaks

Consecutive solved games

1.Sonar Reasoning43 games
2.Sonar Reasoning Pro21 games
3.Deepseek R115 games
4.GPT-4.15 games
5.Llama 4 Maverick4 games

Latest Games

View all games →

#703 - 2025-05-14

Difficulty: 3.3
1 / 23 solvedBest: 19.3s (Pplxsonarreasoning)

#702 - 2025-05-13

Difficulty: 2.8
3 / 24 solvedBest: 9.3s (Pplxsonarreasoning)

#701 - 2025-05-12

Difficulty: 3.0
2 / 23 solvedBest: 17.6s (Pplxsonarreasoning)

Site Status:

Having some issues with Gemini 2.5 returning results. Working on a solution.