No — if you have heard the DeepSeek V4 Pro API discontinued rumour, DeepSeek's own API documentation says the opposite: "In response to user demand, we have decided to continue providing API services for DeepSeek V4 Pro after September 14, 2026, with the billing method remaining unchanged." That statement sits on the official DeepSeek API docs (api-docs.deepseek.com, checked 14 September 2026), and it means the retirement date that had builders migrating in a hurry has been scrapped. Your deepseek-v4-pro API calls keep working after today, at the same prices you were already paying.
📺 Watch: How to use DeepSeek V4.1 Flash API for FREE!
🔥 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 →
DeepSeek V4 Pro API Discontinued? What DeepSeek Actually Announced
The confusion is understandable, because DeepSeek genuinely has been retiring things this month — just not V4 Pro. According to the same official docs, the legacy deepseek-v4-flash model name has been retired: requests sent to that old name are now served by the newer DeepSeek-V4.1-Flash model and billed at the standard Flash price. So one V4-era name really did disappear, and it is easy to see how "deepseek-v4-flash is gone" mutated into "the whole V4 line is being shut down" as it passed through social feeds. The docs draw a clean line between the two situations: the Flash tier was upgraded in place under a new model, while V4 Pro — the heavyweight of the line — keeps its own API service and its own billing "in response to user demand".
Two practical facts follow from that wording. First, there is no forced migration: code pointed at deepseek-v4-pro does not need touching, today or on any announced date. Second, DeepSeek has not published a new end-of-life date to replace the old one — the continuation is open-ended as of 14 September 2026, which is different from a guarantee. If your product depends on V4 Pro, the sensible read is that you have breathing room, not permanence, and the model naming shuffle on the Flash side shows how quickly DeepSeek is willing to move names underneath you.
If you want working AI systems built on models like DeepSeek without burning weeks on trial and error, the AI Profit Boardroom hands you the full Agent OS, 1,000+ done-for-you workflows and five live coaching calls a week → see what is inside the Boardroom. Want tailored advice before you commit? Book a free SEO strategy session and talk it through 1-on-1.
What DeepSeek V4 Pro Costs Now the API Is Staying
Since billing is "remaining unchanged", the current published prices are the ones that carry forward. Per the official DeepSeek pricing page (checked 14 September 2026), DeepSeek-V4-Pro offers a 1 million token context window with time-of-day pricing: peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, and every other hour is off-peak. The gap between the two bands is a straight 50% — peak is exactly double off-peak across every line item.
| DeepSeek API (per 1M tokens) | V4 Pro off-peak | V4 Pro peak | Flash off-peak | Flash peak |
|---|---|---|---|---|
| Input — cache hit | $0.022 | $0.044 | $0.003 | $0.006 |
| Input — cache miss | $0.66 | $1.32 | $0.15 | $0.30 |
| Output | $1.98 | $3.96 | $0.60 | $1.20 |
Both tiers carry the 1M-token context window, and the Flash tier (the deepseek-flash model ID, now served by V4.1-Flash) additionally lists a 384K max output. The pricing spread tells you how DeepSeek positions the pair: Flash runs at roughly a quarter to a fifth of V4 Pro's cost line for line, so V4 Pro only earns its keep on work where the stronger model measurably changes the output — long-context analysis, harder reasoning chains, agent runs where a wrong step costs more than the tokens do. For high-volume, tool-heavy agent loops, Flash pricing is the number that matters, especially that $0.003 off-peak cache-hit rate, which makes repeated-context agent work close to free.
Three Moves That Cut Your DeepSeek V4 Pro Bill
The published price table gives you three levers before you ever touch your prompts. First, scheduling: because peak is limited to 01:00–04:00 and 06:00–10:00 UTC on weekdays, batch jobs shifted outside those windows pay half price on every token — and weekends are entirely off-peak. Second, caching: the cache-hit input rate is 30 times cheaper than cache-miss on V4 Pro ($0.022 versus $0.66 off-peak), so structuring agents to reuse a stable system prompt and context prefix is worth far more here than on flat-rate APIs. Third, tier selection: with the legacy deepseek-v4-flash name retired, requests to it are billed at Flash prices anyway, so there is no cost reason to cling to old model names — pin the current IDs and route only genuinely hard calls to V4 Pro.
If you are running DeepSeek as the brain behind an agent rather than calling the API raw, those same levers apply one layer up. The DeepSeek harness guide covers the agent wrapper around the API, the DeepSeek expert mode write-up covers pushing the model on harder reasoning work, and the Hermes + DeepSeek setup shows the pairing most Boardroom members run for automation. For local and hybrid setups, DeepSeek V4 on Ollama walks through taking the API out of the loop entirely for the workloads where a local model is enough.
A Worked Example on the Published Rates
Put real numbers on it. Say your agent pipeline pushes 10 million input tokens and 1 million output tokens through V4 Pro in a day, with a typical 80% prompt-cache hit rate. Off-peak, that is 8M cached-input tokens at $0.022 ($0.18), 2M cache-miss tokens at $0.66 ($1.32) and 1M output tokens at $1.98 — about $3.48 for the day. Run the identical workload inside the weekday peak windows and it doubles to roughly $6.95. Now route the same job to Flash off-peak: 8M cached tokens at $0.003 ($0.02), 2M cache-miss at $0.15 ($0.30) and 1M output at $0.60 — about $0.92. Same tokens, three very different bills, and all three configurations are just scheduling and routing decisions on prices DeepSeek publishes openly. That is why the discontinuation scare mattered so much: at these rates, V4 Pro is one of the cheapest frontier-class APIs in production use, and the businesses built on that arithmetic wanted certainty it would still exist on 15 September.
How to Verify DeepSeek Announcements Yourself
Deprecation rumours around DeepSeek models spread faster than the documentation that settles them, so it is worth knowing where the authoritative answers live. The news and pricing sections of DeepSeek's API documentation are the primary source — the V4 Pro continuation statement quoted above sits there verbatim, alongside the note that the legacy deepseek-v4-flash name now resolves to V4.1-Flash at Flash pricing. If a headline claims a DeepSeek model is being shut down, check whether the docs list an end-of-life date for the specific model ID you call; a retired alias and a retired service look identical in a tweet and completely different in the documentation. Thirty seconds at the source beats a weekend of unnecessary migration work.
📺 Watch: Deepseek v4.1 is SCARY GOOD!
What This Means for the Rest of the V4 Line
Today's date matters because 14 September 2026 was the pivot DeepSeek's docs anchor everything to: V4 Pro service confirmed to continue past it, and the V4-era Flash naming already folded into V4.1-Flash before it. Read together, the announcements sketch the roadmap — DeepSeek upgrades the cheap tier aggressively and in place, while the premium tier persists on demand rather than on a schedule. If you want the capability side of that story, the DeepSeek V4.1 Flash vision capabilities breakdown covers what the new Flash generation actually does, and the DeepSeek V4 tutorial remains the practical starting point for the model family as a whole. For how DeepSeek-powered stacks compare against other agent brains on real tasks, the Goldie Bench write-up tracks those head-to-head results, and the Agent OS guide covers the operating layer you would run any of them under.
The bottom line for planning: nothing in the official documentation suggests an imminent shutdown, the billing method is explicitly unchanged, and the only deprecation actually executed this month was a model-name consolidation on the Flash tier. Keep your V4 Pro integrations as they are, pin current model IDs rather than legacy names, and treat the peak/off-peak windows and cache-hit rates as the real cost story — because on DeepSeek's published numbers, when you call the model matters almost as much as which model you call. The builders who win on DeepSeek are not the ones who guessed right about the deprecation rumour — they are the ones who read the pricing page closely enough that the rumour never changed their plans in the first place.
If you want to turn cheap frontier models like DeepSeek into systems that actually make money, the AI Profit Boardroom is where the workflows, prompt libraries and daily tutorials live → get inside the Boardroom. Or start with a conversation: book a free SEO strategy session and get a plan for your exact situation.
