DeepSeek V4-Pro Quietly Appears in API Docs, Official Silent
On the evening of August 12, the DeepSeek official API documentation added the DeepSeek-V4-Pro-0813 model, supporting both thinking and non-thinking modes. However, the official team has not yet published a changelog. Previously, the official team promised the V4-Pro release version would be launched as soon as possible—is this a late-night surprise or an early leak?
On the evening of August 12, a new name appeared on the API documentation page of DeepSeek's official website: **DeepSeek-V4-Pro-0813**.
On the Models & Pricing page, this version is now selectable, supporting both non-thinking and thinking modes.
As of the time of writing, the official team has not released any changelog.
This is a bit interesting. On July 31, the official DeepSeek-V4-Flash API went live for public beta, and at the time the official team said the V4-Pro release version would be launched as soon as possible.
Half a month later, the Pro version with build number 0813 appears in the docs, but the official team remains silent.
Based on past practices, DeepSeek tends not to announce releases in advance. Last time, V4-Flash went live directly, and the community discovered it on its own. This time, the Pro release follows a similar pattern, except even an announcement hasn't caught up yet.
For now, the only confirmed details are these: the model version number and the supported modes. As for performance parameters, context length, and pricing, the page doesn't include any additional notes.
If you spot this model in the API docs, you can give it a try. But note that until an official changelog is published, you should hold off on switching your production environment.
Related reading:
- [DeepSeek-V4-Flash Release API Goes Live for Public Beta, V4-Pro Release Version to Launch "Soon"](https://www.ithome.com/0/984/116.htm)
发布时间: 2026-08-12 23:53