On August 13, 2026, DeepSeek officially released the official version of V4 Pro, with the model version updated to DeepSeek-V4-Pro-0813, while the model name for invocation remains unchanged. This version supports a context length of 1M and a maximum output of 384K, offering both thinking and non-thinking modes. It enhances Agent and Coding capabilities and supports Responses API and Codex integration. The current pricing is as follows: input with cache hit is 0.025 yuan/million tokens, input without cache hit is 3 yuan/million tokens, and output is 6 yuan/million tokens.