Side-by-side benchmark comparison across knowledge, coding, math, and reasoning.
DeepSeekMath V2 wins overall with a score of 64 vs 48 (16 point difference).DeepSeekMath V2 wins 4 out of 4 categories.
DeepSeekMath V2
77.8
Llama 3 70B
56.5
DeepSeekMath V2
72
Llama 3 70B
50
DeepSeekMath V2
79
Llama 3 70B
57
DeepSeekMath V2
76
Llama 3 70B
55
DeepSeekMath V2 scores higher overall with 64 vs 48, a difference of 16 points across all benchmarks.
DeepSeekMath V2 leads in knowledge tasks with an average score of 77.8 vs 56.5.
DeepSeekMath V2 leads in coding with an average score of 72 vs 50.
DeepSeekMath V2 leads in math with an average score of 79 vs 57.
DeepSeekMath V2 leads in reasoning with an average score of 76 vs 55.