DeepSeek · 2026-09-08 · major
DeepSeek V4.1 Flash — a two-day beta of a natively multimodal rebuild
DeepSeek V4.1 Flash opened as a limited beta on September 8. DeepSeek describes it as an interim model on a new architecture with multimodal input built in, priced like deepseek-v4-flash and capped at 20 concurrent requests per account.
DeepSeek opened a two-day public test of an interim Flash model rebuilt to take images as a first-class input.
Quick facts
| Maker | DeepSeek |
|---|---|
| Model name | deepseek-v4.1-flash-expires-on-0910 |
| Status | Limited beta, goes offline September 10, 2026 |
| Input | Text and images, handled natively |
| Price | Same as deepseek-v4-flash during the beta |
| Rate limit | 20 concurrent requests per account |
| Access | DeepSeek API, base_url unchanged |
What is it?
DeepSeek V4.1 Flash arrived on September 8 as a limited beta rather than a normal launch, reachable by pointing the existing DeepSeek API at the model name deepseek-v4.1-flash-expires-on-0910. DeepSeek calls it an interim version between the V4 Flash and V4 Pro lines and says it is stronger, faster and cheaper to run than what is live today.
How does it work?
The change under the hood is a new model architecture in which multimodal input is part of the base design instead of a bolted-on extension. That is the split from DeepSeek-V4-Flash-Vision-Exp, the August experiment that taught the existing V4 Flash to accept images. Beta traffic keeps the same base_url and is limited to 20 concurrent requests per account.
Why does it matter?
A beta that expires in two days tells you what DeepSeek is testing rather than shipping, and the survey it sent with the endpoint asks the loudest question directly: can a Flash-priced model replace DeepSeek V4 Pro? If the answer holds up, the cheap tier absorbs work that today needs the expensive one, which is a pricing story as much as a modelling one.
Who is it for?
developers on the DeepSeek API and teams tracking Chinese frontier labs
Frequently asked questions
- How much does DeepSeek V4.1 Flash cost during the beta?
- DeepSeek V4.1 Flash bills at exactly the same rates as deepseek-v4-flash for the length of the beta. DeepSeek did not publish a separate price sheet for the interim model, so anyone already budgeting for V4 Flash can test V4.1 Flash without changing their cost model. The company has not said what the model will cost after general release.
- Why does the DeepSeek V4.1 Flash model name end in expires-on-0910?
- The beta identifier deepseek-v4.1-flash-expires-on-0910 stops working on September 10, 2026. DeepSeek built the expiry into the string so developers can see at a glance that this is a temporary test endpoint rather than a stable model alias. Code that hardcodes the name will start failing on that date rather than silently falling back.
- How is DeepSeek V4.1 Flash different from DeepSeek-V4-Flash-Vision-Exp?
- DeepSeek V4.1 Flash is built on a new model architecture that handles multiple input types as part of the base design. DeepSeek-V4-Flash-Vision-Exp, released in August, instead added image input on top of the existing V4 Flash model as an experimental extension. The V4.1 rebuild is DeepSeek's attempt to stop treating vision as an add-on.
- Can DeepSeek V4.1 Flash handle production traffic?
- No. DeepSeek caps DeepSeek V4.1 Flash at 20 concurrent requests per account and shuts the endpoint down on September 10, which rules out serving real users from it. DeepSeek also circulated a feedback survey with the beta asking testers whether V4.1 Flash could stand in for DeepSeek V4 Pro, so the run is a validation exercise.
Try it
model="deepseek-v4.1-flash-expires-on-0910" against https://api.deepseek.com