Latest release: DeepSeek V4.1-Flash-1017
V4.1 Flash updated to V4.1-Flash-1017
DeepSeek's fast, low-cost model got a significant upgrade. It's much better at multi-step project development, finishes tasks with fewer tokens, so they cost less, and responds faster. The same release added web search and fetch to the API, so apps can answer from live web pages.
Read the full release notes →| Change | In one line | Learn more |
|---|---|---|
| Project development | Better at multi-file, multi-step work such as features, refactors and wiring up APIs | Harness Coding |
| Lower cost per task | Fewer tokens and fewer retries for the same result | API pricing |
| Faster responses | Quicker first reply and quicker complete answers | V4.1 Flash guide |
| Web search - fetch | The API can search the web and read pages before it answers | Search - fetch guide |
Release log
Major DeepSeek releases, newest first. Filter by area:
-
V4.1-Flash-1017
Upgraded Flash build: stronger project development, lower cost, faster responses. Existing API integrations get it automatically, with no code change.
-
Web search - fetch tools in the API
Built-in tools let a request search the web and read a URL before answering. You opt in per request, and the results are billed as tokens.
-
Harness Desktop: public preview, worldwide and open source
DeepSeek's desktop agent is available to everyone. It handles everyday tasks, coding, documents, spreadsheets and scheduled tasks, with composable plugins that extend what it can do.
-
V4.1 Flash
The fast, low-cost generation that became the default model in the chat app and the cheapest option in the API.
-
V4 Pro
The flagship general model, for long, nuanced and high-stakes work. See how it compares with Flash.
-
R1 reasoning model and DeepThink
An open-weight reasoning model that shows its chain of thought. It powers the DeepThink toggle in the chat app, and distilled versions are available for self-hosting.
-
Coder
A code-specialised model family in several sizes, with fill-in-the-middle support for IDE completion.
-
DeepSeek mobile apps for iOS and Android
Free chat on your phone with DeepThink, Search, file uploads and voice input, synced with the web.
No releases in this category yet.
What the latest updates mean for you
| If you are… | What changed for you | Suggested next step |
|---|---|---|
| A chat app user | Default answers are faster, and Flash handles bigger tasks better. | Nothing to do. Try a harder request. How to use DeepSeek |
| An API developer | Lower cost per task, and web search - fetch are now available. | Review your costs on pricing, and try search - fetch. |
| A Harness user | Flash-based sessions finish multi-file work faster and more cheaply. | Use Flash for most sessions. Harness guide |
| Choosing between models | Flash closes more of the gap with V4 Pro on everyday work. | Re-test your prompts. V4 Pro vs V4.1 Flash |
| Comparing with other AI tools | DeepSeek's price advantage has grown. | See DeepSeek vs ChatGPT. |
Do I need to do anything after an update?
| Type of update | Action needed? | What to check |
|---|---|---|
| New build of an existing model (like 1017) | Usually no. The model name stays the same and you get the new build automatically. | Re-run your key prompts or tests. Output style can shift slightly. |
| New API feature (like search - fetch) | Optional. It's off until you add it to your requests. | The feature's guide, and how it's billed. |
| Price change | Review your budget. | Your Usage page and the pricing page. |
| Model name change or retirement | Yes. Update the model field before the old name stops working. | Search your code for the old name. See the API guide. |
| App update (mobile or Harness) | Update the app. | Install only from official links on our download page. |
| Outage or "server busy" | Wait and retry. | Use retries that wait longer each time. See API errors or the login - troubleshooting guide. |
How to stay up to date
Bookmark this page
We add every major release with a plain-English summary and a "do I need to act?" note.
API platform notices
Changes to models, prices and limits appear in the console on the API platform.
DeepSeek updates - frequently asked questions
What is the latest DeepSeek release?
V4.1-Flash-1017, an upgraded build of V4.1 Flash that also brought web search and fetch to the API. Release notes.
Will an update break my API integration?
New builds of an existing model keep the same name and request format, so they shouldn't. Re-test important prompts anyway, because output style can change slightly.
How do I know which model version I'm using?
Every API response includes a model field. In the chat app, the default is V4.1 Flash, and DeepThink uses R1.
Do chat app users get updates automatically?
Yes. Model upgrades apply automatically on the web. For the mobile apps, install new versions from the official stores.
Where can I compare all the current models?
On DeepSeek models explained, which compares V4 Pro, V4.1 Flash, R1 and Coder.
Is this DeepSeek's official changelog?
No. DeepSeekR1.com is an independent guide. Official announcements come from DeepSeek. See about us, our disclaimer, privacy policy and terms.