DeepSeek V4.1 Flash 1017 vs Kimi K3.1 Pro

Each Map
DeepSeek V4.1 Flash 1017DeepSeek · reasoning medium	vs	Kimi K3.1 ProMoonshot AI · reasoning high
30.08th of 12	Crusader Bench ScoreMean of the map scores	32.15th of 12
Kimi K3.1 Pro scores 2.1 higher over every map, 32.1 to 30.0, and higher on two of the three maps.
27.210th of 12		40.94th of 12
39.65th of 12		17.88th of 12
23.311th of 12		37.64th of 12
Score Breakdown
29%	Net worth growth80% of the score	27%
29%	Population growth10% of the score	40%
41	Popularity10% of the score	63
Cost and Effort, One Game per Map
6	Games	6
$0.39	CostLower is better	~$3.40
6.6M	Tokens	8.7M
131	Turns	168
1 h 55 min	Wall-clock time	2 h 20 min

Ranks are among the models with a full-budget score; over every map, among those with one on every map. Net worth and population growth are each the share of a person’s gain over doing nothing. Cost and effort add up one game on each map, the mean of its games where a model played a map more than once.