Overview
DeepSeek V4.1 Flash opened as a limited public beta on 8 September 2026. It is reachable through the existing DeepSeek API — the base_url does not change — by setting the model name to `deepseek-v4.1-flash-expires-on-0910`. The expiry is baked into the identifier: the endpoint is scheduled to go offline on 10 September 2026, so this is a two-day test window rather than a general release.
DeepSeek describes V4.1 Flash as an interim version sitting between the V4 Flash and V4 Pro tiers, built on a new model architecture in which multimodal input is part of the base design. That is the difference from DeepSeek-V4-Flash-Vision-Exp, the August experiment that added image input on top of the existing V4 Flash model as an extension. DeepSeek says the new model is stronger, faster to generate and cheaper to run than what is live today.
During the beta, billing matches the deepseek-v4-flash tariff and each account is capped at 20 concurrent requests, a limit that rules the endpoint out for production traffic. DeepSeek circulated a feedback survey alongside the beta asking testers whether the V4.1 Flash interim version could fully replace the live DeepSeek V4 Pro — which is the question the run exists to answer. DeepSeek has not published a model card, benchmark table or separate price sheet for V4.1 Flash.
| Released | 2026-09-08 |
|---|---|
| License | Undisclosed |
| Weights | API only |
| Modalities | Text, Vision |
| Status | Limited-beta preview on the DeepSeek API Platform under the model id `deepseek-v4.1-flash-expires-on-0910`; the endpoint goes offline on 10 September 2026. Not generally available. |
Strengths
- Multimodal input is handled by the base architecture rather than by an extension bolted onto a text model
- Drop-in for existing DeepSeek API users: the base_url is unchanged and only the model name differs
- Billed at DeepSeek-V4-Flash rates for the length of the beta, so testing costs no more than current Flash traffic
- DeepSeek positions it as Flash-tier speed and price with capability aimed at the DeepSeek-V4-Pro tier
Best for
- Testing whether Flash-priced inference can absorb work currently sent to DeepSeek V4 Pro
- Evaluating mixed text-and-image prompts on a DeepSeek model whose architecture handles both natively
- Benchmarking a new DeepSeek architecture before it reaches general availability
How to access
| Provider | Model ID |
|---|---|
| DeepSeek Platform ↗ | deepseek-v4.1-flash-expires-on-0910 |
DeepSeek V4 — every version
The full lineage of the DeepSeek V4 line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.
| Version | Released | Context | License |
|---|---|---|---|
| DeepSeek V4.1 Flash | 2026-09-08 | — | Undisclosed |
| DeepSeek-V4-Flash-Vision-Exp | 2026-08-21 | 1M | MIT |
| DeepSeek-V4-Procurrent | 2026-08-13 | 1M | MIT |
| DeepSeek-V4-Flash | 2026-07-31 | — | MIT |
FAQ
Is DeepSeek V4.1 Flash generally available?
No. DeepSeek V4.1 Flash opened on 8 September 2026 as a limited beta under the model id `deepseek-v4.1-flash-expires-on-0910`, with the endpoint scheduled to go offline on 10 September 2026 and each account capped at 20 concurrent requests. DeepSeek has not published a model card, benchmark table or standalone price sheet for it.
How does DeepSeek V4.1 Flash differ from DeepSeek-V4-Flash-Vision-Exp?
DeepSeek V4.1 Flash is built on a new model architecture where multimodal input is part of the base design. DeepSeek-V4-Flash-Vision-Exp, released on 21 August 2026, instead added image input to the existing DeepSeek-V4-Flash model as an experimental extension. The V4.1 rebuild is DeepSeek's move away from treating vision as an add-on.
What does DeepSeek V4.1 Flash cost?
During the beta, DeepSeek V4.1 Flash bills at the same rates as deepseek-v4-flash. DeepSeek did not publish a separate price sheet for the interim model and has not stated what it will cost if it reaches general availability, so anyone already budgeting for V4 Flash traffic can test it without changing their cost model.