Scores 54.1 on the Extended NYT Connections Benchmark, a large improvement over Gemini 2.0 Flash Thinking Experimental 01-21 (23.1).

1 o1-pro (medium reasoning) 82.3

2 o1 (medium reasoning) 70.8

3 o3-mini-high 61.4

4 Gemini 2.5 Pro Exp 03-25 54.1

5 o3-mini (medium reasoning) 53.6

6 DeepSeek R1 38.6

7 GPT-4.5 Preview 34.2

8 Claude 3.7 Sonnet Thinking 16K 33.6

9 Qwen QwQ-32B 16K 31.4

10 o1-mini 27.0

https://github.com/lechmazur/nyt-connections/