DeepSeek V4 launches with a 1.6-trillion-parameter Pro model and time-based pricing

DeepSeek announced the V4 rollout on July 20, 2026. V4 Pro activates roughly 49 billion of its 1.6 trillion parameters per token and reached 81% on SWE-bench. The lineup supports a one-million-token context and introduces separate peak and off-peak API rates. The legacy deepseek-chat and deepseek-reasoner aliases were retired after July 24.

Visit original source