Page 12 — connections-bench

Recent puzzle days and how each language model performed. Pick a date for the full board and guess-by-guess replay. See About for more info and how it works.

Sun, Aug 2, 2026
- claude-haiku-4.5Lost2/4 categories4 mistakes7 guesses
- gemini-3.7-flashSolved0 mistakes4 guesses
- gpt-4.1Lost2/4 categories4 mistakes6 guesses
- gpt-4.1-miniLost2/4 categories4 mistakes7 guesses
No selected models for this date
Sat, Aug 1, 2026
- claude-haiku-4.5Lost2/4 categories4 mistakes6 guesses
- gemini-3.7-flashSolved0 mistakes4 guesses
- gpt-4.1Lost2/4 categories4 mistakes7 guesses
- gpt-4.1-miniLost2/4 categories4 mistakes24 guesses
No selected models for this date

Season standings

Model	Win rate	Avg mistakes	Avg Game Time	Cost/Game	Tokens/Game	Games
glm-5.3	100%4W / 0L	0.0	285s	$0.116	28k	4
gpt-5.6-sol	100%4W / 0L	0.0	17s	$0.0099	2.9k	4
grok-4.6	100%4W / 0L	0.0	203s	$0.067	14k	4
claude-opus-5	100%3W / 0L	0.0	22s	$0.052	6.1k	3
deepseek-v4-flash-0731	100%48W / 0L	0.0	464s	$0.0076	41k	48
gemini-3.7-flash	100%79W / 0L	0.0	19s	$0.0084	4.4k	79
gemini-3.8-flash	100%48W / 0L	0.1	20s	$0.0069	4.1k	48
qwen3.8-flash	100%5W / 0L	0.2	933s	$0.023	54k	5
gpt-4.1-nano	100%4W / 0L	1.3	6.8s	$0.0005	4.3k	4
glm-5.3-flash	98%62W / 1L	0.3	262s	$0.0053	13k	63
claude-sonnet-5	97%63W / 2L	0.2	43s	$0.039	8.2k	65
gpt-5.6-luna	94%45W / 3L	0.6	31s	$0.0027	4.8k	48
grok-4.3	92%58W / 5L	0.4	95s	$0.018	9.8k	63
gemini-3.5-flash-lite	84%54W / 10L	1.2	6.6s	$0.0026	5.0k	64
claude-haiku-4.5	81%64W / 15L	1.0	48s	$0.032	10k	79
gpt-4.1	73%58W / 21L	1.4	8.5s	$0.011	4.7k	79
gpt-4.1-mini	48%38W / 41L	2.4	14s	$0.0024	6.0k	79