DeepSeek updates the deepseek-v4-pro API to DeepSeek-V4-Pro-0813
DeepSeek official documentation confirms that deepseek-v4-pro now serves DeepSeek-V4-Pro-0813 under the unchanged API model name, with a one-million-token context window, up to 384K output tokens, and Responses API support.
What happened
Event details
DeepSeek official Getting Started documentation states that the model version behind deepseek-v4-pro has been updated to DeepSeek-V4-Pro-0813 and that callers can keep using the same model name. The official Models and Pricing page also lists a one-million-token context window, up to 384K output tokens, thinking and non-thinking modes, JSON Output, Tool Calls, the Responses API, and the Anthropic-compatible API. DeepSeek’s official Codex integration guide further confirms native Responses API support and documents using Pro and Flash with Codex CLI, the ChatGPT desktop app, and the Codex extension for VS Code. The guide belongs to the same August 12 documentation build as the version switch, rather than a second release afterward.\n\nThe event date is August 12, 2026, the first verifiable update date of the official pages. The 0813 suffix is treated as a version label, not as an independently announced release date. At verification time, the official changelog still ended on July 31, the existing Hugging Face V4-Pro card still described a preview, and DeepSeek had not published separate 0813 weights, a dedicated model card, or benchmarks attributable specifically to 0813.
Assessment
Why it matters
This update moves DeepSeek V4-Pro onto its post-preview API main line and confirms broad developer-interface support, while the lack of a standalone technical release limits what can be attributed specifically to the 0813 version.
Availability
Access notes
Available through the DeepSeek API as deepseek-v4-pro, with an official Codex/Responses integration guide. A standalone 0813 announcement, new weights, and a dedicated model card had not been published at verification time.