DeepSeek V4.1 Flash
has not shipped
As of 2026-09-09 the newest entry in DeepSeek's official changelog is still the 2026-08-21 V4-Flash-Vision-Exp release, and the API docs list only three callable models: deepseek-v4-flash, deepseek-v4-pro and deepseek-v4-flash-vision-exp. There is no V4.1 entry anywhere. Chinese business media reported a launch window around 09-10, but no matching official announcement exists. This page keeps what is official separate from what is only reported.
Updated 2026-09-09
Four checkable facts
Newest official changelog entry
It announces DeepSeek-V4-Flash-Vision-Exp. Nothing has been added since, and nothing mentions V4.1.
Models the official docs list
deepseek-v4-flash, deepseek-v4-pro and deepseek-v4-flash-vision-exp. There is no fourth.
What the two main ids point at
The docs state deepseek-v4-flash was updated to V4-Flash-0731 and deepseek-v4-pro to V4-Pro-0813.
Official mentions of V4.1
We checked the changelog, the models and pricing docs, and the first-API-call page. Not one mention.
What you can actually use now
If you came looking for how to wire up V4.1 Flash, the answer is that you cannot yet: it has not been released and there is no callable model id. What is live is still deepseek-v4-flash (pointing at V4-Flash-0731) and deepseek-v4-pro (pointing at V4-Pro-0813), plus an experimental multimodal deepseek-v4-flash-vision-exp. All three can be called on QCode with one key.
The gap between reporting and announcement
Chinese business and tech outlets reported on 09-08 and 09-09 that V4.1 Flash was in a time-limited beta and planned for release around 09-10, and that once live, requests to deepseek-v4-pro would be routed to V4.1 Flash and billed at the lower rate. Those claims currently exist only in media retellings. We checked DeepSeek's official changelog, the models and pricing docs, and the first-API-call page, and found no matching entry in any of them. The gap itself is information: until an official notice appears, do not plan a migration around that timetable.
By date
Official changelog: deepseek-v4-flash updated to DeepSeek-V4-Flash-0731. That is the actual version behind the flash line today.
Official changelog: deepseek-v4-pro updated. The docs state the id now points at DeepSeek-V4-Pro-0813.
Newest official changelog entry: the multimodal vision model DeepSeek-V4-Flash-Vision-Exp goes live. As of 2026-09-09 it is still the last one.
Official versus reported
✅ Officially confirmed
DeepSeek's docs currently list three callable models — deepseek-v4-flash, deepseek-v4-pro and deepseek-v4-flash-vision-exp — with the first two pointing at V4-Flash-0731 and V4-Pro-0813, and the newest changelog entry dated 2026-08-21. Every one of those can be checked line by line against the official documentation.
⚠️ Rumour: media retelling only, unconfirmed by the vendor
The time-limited beta codename, the launch window around 09-10, the new price list, and the claim that v4-pro requests will be rerouted to V4.1 Flash and billed at the Flash rate all appear only in media reports. None has surfaced through an official channel and no official model id has been published. This page does not adopt those numbers; it only records that they exist. If DeepSeek ships, we will rewrite this page from the official documentation.
Two ways to wait
Wait for V4.1
If your system pins deepseek-v4-pro and the reported rerouting is real, your billing basis changes when it happens. The risk is that the wait is open-ended: there is no official date.
Benchmark the two live tiers now
deepseek-v4-flash and deepseek-v4-pro are both available. Run the same real requests against each and build your own cost and quality baseline, so the next version arrives with a control group already in place.
Where to watch
Three places are enough: the official changelog (new models get an entry there), the models list in the API docs (callable ids are defined there), and the pricing page (rate changes land there). Do not treat a beta alias that appears on a third-party gateway as authoritative — it is not an official id and there is no guarantee it matches the final release. This page will be rewritten from the official docs once an entry appears.
On QCode
deepseek-v4-flash and deepseek-v4-pro can both be called on QCode with a single key, and both have real traffic over the last 30 days. V4.1 Flash has not been released, so no platform can offer it and neither can QCode. Once it ships, connecting will work like any existing model: same endpoint, same key, change the model field.
Frequently asked questions
Has DeepSeek V4.1 Flash been released?
As of 2026-09-09 there is no official release announcement. The official changelog stops at 2026-08-21 and the API docs list only v4-flash, v4-pro and v4-flash-vision-exp.
So is the 09-10 launch news false?
Not necessarily. Chinese media did report that window, but no official announcement has appeared. The correct reading is that this is reporting, not a vendor commitment. DeepSeek has had reported dates diverge from actual releases before, so planning a migration around the report carries risk.
What will the official model id be?
Not published. It could be a new string, or an in-place upgrade of the existing deepseek-v4-flash — DeepSeek has done both. Until it appears in the official docs, any specific id is a guess.
My system pins deepseek-v4-pro. Should I worry?
Worth watching, but nothing to change today. The reports say v4-pro requests will be rerouted and billed at the lower rate once the new model is live; if true, that is a behaviour and billing change rather than an interface change. Re-assess when an official notice appears, and validate the response shape on a small slice of traffic first.
Should I be on flash or pro right now?
It depends on the task. The pro line is stronger but costs more; the flash line is fast and cheap. The practical move is to run your own real requests against both ids and read the results rather than copying someone else's conclusion. On QCode both share one key and one quota, so switching is a model-field change.
Will this page be updated after the release?
Yes. The whole structure of this page is to keep official statements separate from media reports; once an official entry appears we will rewrite it from the documentation and keep a record of this period.
Sources
The official changelog, the API models and pricing documentation, and the first-API-call page, all captured 2026-09-09. None of the three contains a V4.1 entry — that is a negative finding, and we checked each page rather than inferring it. Media reports are recorded as existing; their numbers are not adopted here. QCode call volumes come from this site's own 30-day usage statistics.
Build your baseline on the two live tiers first
deepseek-v4-flash and deepseek-v4-pro share one QCode key and one quota with Claude, GPT and Gemini; switching is a model-id change. Plans start at ¥60/month and the key is live the moment you pay.
Related reading
DeepSeek V4 Pro 0813 guide
The actual version behind the pro line today: positioning, performance and how to connect.
DeepSeek V4 Flash 0731 guide
The actual version behind the flash line: speed, price and where it fits.
DeepSeek V5 tracker
The other repeatedly rumoured line, and what the vendor has actually said about it.
This page separates officially confirmed information from media reports; the latter has not been confirmed by DeepSeek and is not a release commitment or timetable. All official information was captured 2026-09-09 and may be updated at any time — check DeepSeek's official documentation.