math Benchmark

2 models evaluated

#	Model	Provider	Type	Score
1	DeepSeek V4 Pro Base	DeepSeek	Open	64.5%
2	DeepSeek V4 Flash Base	DeepSeek	Open	57.4%