The DeepSeek V4.1 Flash release date is 10 September 2026, Beijing time: DeepSeek announced on its open platform on 9 September 2026 that the new model would be officially released around that date, and that from launch every request sent to V4 Pro will be answered by V4.1 Flash and billed at V4.1 Flash unit pricing until V4.1 Pro arrives. So this is not just another budget model quietly joining the price list — the current flagship's traffic is being folded into the new release from day one, which makes this the most consequential DeepSeek launch since V4 Pro went to general availability in August.
📺 Watch: DeepSeek Just Made V4 Flash 10X Better
🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside · Want AI SEO help 1-on-1? Book a free SEO strategy session →
One note on sourcing before the detail. Everything factual on this page comes from DeepSeek's launch announcement of 9 September 2026, posted to its open platform and relayed by the Odaily newswire the same day; where older V4 numbers appear, they are attributed to DeepSeek's earlier release notes and API changelog. Fresh model launches attract speculation fast — this page sticks to what the company actually said, and flags clearly where a question is still open.
DeepSeek V4.1 Flash Release Date and What Was Announced
DeepSeek's announcement, posted on 9 September 2026, said V4.1 Flash would be officially released around 10 September 2026, Beijing time. Beijing runs eight hours ahead of UTC, so in practice the launch window opens during the late evening of 9 September for the US and the early hours of 10 September for the UK and Europe. If you are reading this on launch day, the sensible first check is DeepSeek's own platform changelog, since "around" gives the company a little room on the exact hour.
The announcement itself makes four concrete claims about the model. Per DeepSeek, V4.1 Flash:
- has fully surpassed V4 Pro in metrics including performance, cost, speed and total processing time, following internal and external testing across multiple dimensions;
- is built on a new model structure rather than a tweak of the V4 architecture;
- ships with native multimodal support;
- and delivers stronger capability at faster speed and lower cost than the outgoing generation.
Treat those as vendor claims for now, because that is what they are: DeepSeek's own testing, not independent benchmarks. Third-party numbers usually land within days of a DeepSeek release, and the claim worth scrutinising most is the first one — a Flash-tier model that beats the company's own Pro flagship across the board would be a genuinely unusual result, which is presumably why DeepSeek is confident enough to route Pro traffic straight to it.
V4 Pro Requests Route to V4.1 Flash — at Flash Pricing
The most practical line in the announcement is the routing arrangement. After V4.1 Flash officially launches, and before V4.1 Pro is released, all requests sent to V4 Pro will be routed to V4.1 Flash and billed at V4.1 Flash unit pricing. If your agents point at the V4 Pro model name today, they will start getting V4.1 Flash answers from launch day without you changing a line of configuration — and your bill drops to the Flash rate while that arrangement holds.
For context on what V4 Pro billing looked like until now: per DeepSeek's API changelog, V4 Pro reached general availability on 13 August 2026, and from 16 August the API moved to peak and off-peak billing, with off-peak set at half the peak rate and V4 Pro output topping out at 3.96 dollars per million tokens at peak. The full breakdown, including the exact peak windows, is in the DeepSeek V4 pricing update write-up. What the new announcement did not include is a V4.1 Flash price card — so read "billed at V4.1 Flash unit pricing" as the commitment, and check the official pricing page on launch day for the actual per-token numbers before you update any cost models.
If you want to turn model drops like V4.1 Flash into income instead of just headlines you skim, check out the AI Profit Boardroom → get the launch-day agent playbooks. Prefer a 1-on-1 route map for your niche? Book a free SEO strategy session and get a plan built around your site.
Two watch-outs follow directly from the routing. First, if you have evaluations, prompts or output parsers tuned specifically to V4 Pro's behaviour, re-run them on launch day, because the model answering those requests changes even though the endpoint does not. Second, if a workflow genuinely needs the old V4 Pro behaviour, the announcement offers no opt-out — the routing applies to all V4 Pro requests — so the realistic plan is to validate V4.1 Flash quickly rather than try to hold the line.
Where V4.1 Flash Fits in DeepSeek's Flash Lineage
Flash has always been DeepSeek's high-volume workhorse tier: the cheap, fast model you run thousands of times a day for research sweeps, first drafts, sorting, tagging and monitoring, while a pricier model handles the hard thinking. The DeepSeek V4 Flash harness guide covers that division of labour in depth — Flash was never built to win an IQ contest, it was built to be good enough thousands of times in a row for pennies. If V4.1 Flash really does clear V4 Pro on quality while keeping Flash economics, that workhorse tier suddenly gets a much bigger job description.
The V4 generation has moved quickly all year: the V4 preview arrived in April 2026, V4 Pro went GA in August, and late August brought an experimental vision model covered in the V4-Flash-Vision-Exp write-up. Against that backdrop, "native multimodal support" in V4.1 Flash reads like the vision experiment graduating into the main line — though the announcement does not say that explicitly, so treat it as context rather than confirmation until the model card is public.
On the benchmark side, the bar V4.1 Flash has to clear is well documented: DeepSeek's own V4 Pro release notes posted 87.9 on Terminal Bench 2.1, 62.7 on DeepSWE, and 42.7 without tools (60.0 with tools) on Humanity's Last Exam. "Fully surpassed V4 Pro" means beating those numbers at Flash-tier cost. For how the funnel compares agent brains across this class of model in practice, the Goldie Bench write-up covers the testing approach and standings.
📺 Watch: DeepSeek V4 FULL COURSE 6 HOURS (Build & Automate Anything)
What to Do on Launch Day
If you already run DeepSeek in production, nothing breaks by default — the routing is DeepSeek's job, not yours. The launch-day checklist is short: confirm the V4.1 Flash rates on the official pricing page, re-run your evaluation set against the rerouted endpoint, and note the results, because "surpassed V4 Pro on our tests" and "surpassed V4 Pro on your workload" are different claims. If the quality holds on your tasks, the routing period is effectively a free upgrade with a price cut attached.
If you are new to DeepSeek, launch week is a good moment to start, since you get the newest model at the cheapest tier by default. The DeepSeek V4 tutorial walks through the basic setup, the best harness guide ranks the tooling to run it inside, and if you are weighing it against Anthropic's tooling, the DeepSeek Harness vs Claude Code comparison covers that decision. The general principle stands regardless of model: slot new brains into an operating system you already trust rather than rebuilding your stack every launch week — that is the whole argument of the Agent OS approach, where the model is a swappable part rather than the foundation.
What About DeepSeek V4.1 Pro?
No date. The announcement only says the V4 Pro routing arrangement holds until V4.1 Pro is released — which confirms a V4.1 Pro exists on the roadmap, and confirms nothing else. Any specific V4.1 Pro date you see this week is a guess, because DeepSeek has not published one. The precedent from the V4 cycle suggests the gap between a Flash-tier debut and the Pro follow-up is measured in weeks or months rather than days, but that is pattern-reading, not a commitment from DeepSeek.
The strategic read is more interesting than the date. Launching Flash first, routing flagship traffic to it, and billing at the cheaper rate is a cost-led move: DeepSeek is betting that a model which is cheaper, faster and — by its own testing — better will hold users more effectively than a premium flagship would. For anyone running agent fleets on a budget, that bet works in your favour either way: you get today's best DeepSeek model at the lowest tier price while the flagship race plays out. Keep an eye on the official changelog for the V4.1 Pro announcement, and expect the routing — and the temporary discount that comes with it — to end when it lands.
If you want cheap-model agent stacks that pay for themselves — the prompt libraries, daily tutorials and weekly live coaching are all inside — check out the AI Profit Boardroom → join 3,000+ AI operators. And if you would rather have eyes on your own setup first, book a free SEO strategy session and map it out 1-on-1.
