Updated 2026-11-20
DeepSeek V4 Model Names Explained: Pro, Flash, 0813 and 1017
DeepSeek V4 is a family of models. Pro and Flash identify branches of that family. A suffix such as 0813 identifies a particular release, while the model field in an API request takes a shorter name that can continue to work across releases.
| Model | Version served | Send as model |
|---|---|---|
| DeepSeek V4 Pro | DeepSeek-V4-Pro-0813 | deepseek-v4-pro |
| DeepSeek V4.1 Flash | DeepSeek-V4.1-Flash-1017 | deepseek-flash |
What do 0813 and 1017 tell you?
0813 marks the August 13, 2026 release of V4 Pro. Flash followed a different schedule. DeepSeek introduced V4.1 Flash on September 10 and updated it on October 17. The 1017 suffix singles out that October version. If a benchmark omits the suffix, check its source before attributing the score to the October release.
Which name should I put in my client?
For the official API, set the model field to deepseek-v4-pro for Pro or deepseek-flash for Flash. An older configuration might still use deepseek-v4-flash or deepseek-v4-flash-vision-exp. DeepSeek accepts both aliases and routes their requests to DeepSeek-V4.1-Flash-1017 at the Flash rate. An old alias therefore won’t reproduce the behavior of the retired model.
When recording an evaluation, save the provider, the requested API ID, the date and the version documented on that date. For a local run, record the actual weights repository and revision you downloaded.
FAQ
Does deepseek-v4-pro select V4.1 Pro?
As of November 20, DeepSeek maps that ID to DeepSeek-V4-Pro-0813. Its public API table has no separate V4.1 Pro entry.
What does deepseek-ai/DeepSeek-V4.1-Flash-1017 identify?
It is DeepSeek’s weights repository on Hugging Face. Use the repository name to locate a local download and deepseek-flash when calling the official hosted API.
To check whether DeepSeek has published a newer Pro model or API name, consult the dated release-status page.