DeepSeek-V4.1-Flash-1017 Release Date and DeepSeek's Next Model

As of 21 October 2026, DeepSeek's newest model is DeepSeek-V4.1-Flash-1017, released on Saturday 17 October 2026 on DeepSeek's website, app and API, with the weights published openly the same day. On versus, the prediction Will DeepSeek ship its next Flash model by 30 Nov? has been listed as settled since 12:15 UK time (BST) on 17 October 2026, more than six weeks before its end date.

The new model keeps the V4.1 number: DeepSeek calls it a minor update to V4.1-Flash, its release of 10 September. The only later model named on DeepSeek's own pages is a V4.1-Pro, and no page we checked on 21 October 2026 gives it a date. This guide is for readers in Great Britain and is not advice.

What DeepSeek put out on 17 October

The 17 October 2026 entry in DeepSeek's API change log says V4.1 Flash was upgraded to DeepSeek-V4.1-Flash-1017 and that DeepSeek's official website and app moved to it too, with API prices unchanged. DeepSeek's news post of the same day says it is live on all three.

• On the API. The Models and Pricing page now names DeepSeek-V4.1-Flash-1017 as the version behind deepseek-flash.
• Open weights. The weights are in DeepSeek's Hugging Face repository under the MIT licence. DeepSeek gives no time of day; the repository records its creation at 02:40 UTC on 17 October (03:40 BST).
• What changed, by DeepSeek's account. Higher benchmark scores than V4.1-Flash, such as 42.3 against 31.2 on Terminal-Bench 4.0, with shorter run times and shorter outputs. These are DeepSeek's own figures, not independent tests.

How the prediction's rules read

DeepSeek's own information decides this prediction, and solid press reports can confirm it. In our own words, a release had to pass these tests:

1. Family and tier. A main V-series model with DeepSeek and Flash in its name or model ID. Pro models such as the planned V4.1-Pro are out, and so are side models such as the experimental V4-Flash-Vision-Exp.
2. Newer than V4.1-Flash. DeepSeek must present it as a later model: a V4.2-Flash, a V5-Flash or a later build carrying a four-digit date like DeepSeek-V4-Flash-0731.
3. A new name or date label. An update under an unchanged model ID does not count, nor does V4.1-Flash itself in another form, such as re-uploaded weights, a snapshot or the deepseek-flash name pointing at it.
4. Or one model for both tiers. A single successor without separate tiers also counts if it takes over or runs through the Flash endpoint, or if DeepSeek sells it as its fast, cheap model.
5. Open to everyone in the window. Public access between 22:42 UTC on Monday 14 September 2026 (23:42 BST) and 11:59 pm Eastern Time on Monday 30 November, which is 04:59 GMT on Tuesday 1 December, with New York and the UK both off summer time by then. An open beta or open waiting list is enough; closed or invitation-only access is not. DeepSeek must announce it as public, or make it usable by the public and name it clearly on its official website.

If nothing met these tests inside the window, the rules give No.

Set beside those tests, DeepSeek's pages show this: the 17 October model has DeepSeek and Flash in its name, its date label 1017 is new and written like that of DeepSeek-V4-Flash-0731, and the news post says it is live on the website, app and API. DeepSeek also calls it a minor update and serves it under the existing deepseek-flash name. The versus API has shown the prediction as settled since 11:15 UTC on 17 October 2026 (12:15 BST); our guide to what market resolution means explains settlement.

DeepSeek's V4 line, release by release

Dates come from DeepSeek's change log, news posts and Hugging Face organisation, read on 21 October 2026.

• DeepSeek-V4 preview, 24 April 2026. Preview versions of V4-Pro and V4-Flash reached DeepSeek's chat website and API, with open weights (launch post).
• DeepSeek-V4-Flash-0731, 31 July 2026. The official V4-Flash API release, in public beta; its model card says it supersedes the preview.
• DeepSeek-V4-Pro-0813, 13 August 2026. The general release of V4-Pro on the app, the website and the API (news post, model card).
• DeepSeek-V4-Flash-Vision-Exp, 21 August 2026. An experimental image-reading model on the API (news post), a side model under the rules.
• DeepSeek-V4.1-Flash, 10 September 2026. The smallest model in a new architecture family, with native image understanding and 552 billion parameters (launch post). It went live on the API as deepseek-flash, and V4-Flash and V4-Flash-Vision-Exp were retired.
• DeepSeek-V4.1-Flash-1017, 17 October 2026. The update described above.

In the Flash tier, 98 days passed from the V4 preview to V4-Flash-0731, 41 days from there to V4.1-Flash and 37 days to the 17 October build; DeepSeek publishes no schedule, so these gaps are no guide to the next one. By 21 October 2026 nothing newer had appeared in the change log, the API news list or the Hugging Face organisation, and the news page on deepseek.com still led with 10 September.

What DeepSeek has said about its next model

DeepSeek has named a next model only on the Pro side. Its 10 September post said requests for deepseek-v4-pro would go to V4.1-Flash from 04:00 UTC on 14 September (05:00 BST) until V4.1-Pro launched. The change log entry for the same day says instead that V4 Pro stays on the API after 14 September because users asked for it. On 21 October 2026 the Models and Pricing page still lists DeepSeek-V4-Pro-0813 behind deepseek-v4-pro.

The September post also calls V4.1-Flash the smallest model in a family built to scale to larger models, without naming one, and the 17 October posts say nothing about what comes next.

Where the next release would show up

• Any day, on DeepSeek's own pages. For V4-Flash-0731, V4-Pro-0813, V4.1-Flash and V4.1-Flash-1017, the change log entry and the Hugging Face repository carry the same date, so those are the first places to look.
• The Models and Pricing page. A new version name under deepseek-flash or deepseek-v4-pro would show that a new model has reached the API.
• Whenever V4.1-Pro arrives. It would be DeepSeek's first new Pro model since DeepSeek-V4-Pro-0813 of 13 August.

None of these is dated, and none is a forecast.

Before you trade

A price sits between 0p and 100p and is the market's estimate of a chance. A winning contract settles at 100p and a losing one at 0p, so you can lose the whole amount you put into a position. versus adds a 2% transaction fee on top of the stake and shows it before you confirm, and while a market is open you can sell your position at the current price. How it works follows a position through to settlement.

versus is licensed by the UK Gambling Commission under account 101143, listed on the Commission's public register, and positions settle in ordinary money, not cryptocurrency. Safer Play gives you deposit limits, loss limits, reality checks, time out and self exclusion, alongside GAMSTOP. Our Qwen Max 3.9 release date guide covers a similar Flash prediction for Alibaba.

Frequently asked questions

What is the latest DeepSeek model?

DeepSeek-V4.1-Flash-1017, released on 17 October 2026 and served on the API as deepseek-flash. DeepSeek's Pro model on the API is still DeepSeek-V4-Pro-0813, from 13 August 2026.

Is DeepSeek-V4.1-Flash-1017 a new model or an update?

DeepSeek's model card calls it a minor update to V4.1-Flash, and the API name deepseek-flash did not change. It does have its own open weights and a new date label, 1017.

When will DeepSeek release V4.1-Pro?

DeepSeek has not said. Its 10 September post named V4.1-Pro as still to come, and its change log says V4 Pro remains on the API after 14 September.

Is the versus prediction on DeepSeek's next Flash model still open?

No. The versus API has listed it as settled since 12:15 UK time (BST) on 17 October 2026.

18+ only. Trade responsibly. BeGambleAware.org and GAMSTOP.