The Frontier

This balance point implies the optimal batch size is approximately 300 times the model's sparsity ratio. For DeepSeek's sparsity of 32/256, this yields a batch size around 2000-3000 tokens.

DeepSeek released its V4 model family with a 1 trillion parameter Pro version and a smaller Flash version, both featuring a 1 million token context window.

DeepSeek V4 Pro is priced at $174 per million input tokens and $348 per million output tokens, undercutting Opus 4.6 and GPT-5.4 by more than 75%.