On August 13, 2026, the official DeepSeek API documentation revealed that the deepseek-v4-pro model has been updated to DeepSeek-V4-Pro-0813. The invocation method remains unchanged; users can still call the latest version using deepseek-v4-pro. This model supports both non-thinking and thinking modes, with a context length of up to 1M and a maximum output of 384K Tokens. It is priced three times higher than the V4-Flash version.