Release summary
| Release | V4.1-Flash-1017, an updated build of DeepSeek V4.1 Flash |
|---|---|
| Type | Model upgrade plus a new API feature |
| Headline changes | Significantly better project development · lower cost per task · faster · web search - fetch in the API |
| Applies to | Chat app (web - mobile), API, Harness Desktop |
| Action needed | None for most users. Search - fetch is optional, so add it when you want it |
What's new, in detail
1 · Significantly better project development
Flash is better at multi-step, multi-file work: planning a change, keeping edits consistent across files, running checks and recovering from errors. In practice that means features, refactors and wiring up APIs finish in fewer rounds, especially in agent tools like Harness Coding.
2 · Lower cost per task
Tasks are completed at lower cost. Fewer wasted steps and retries mean fewer tokens for the same result. See current rates on API pricing.
Quicker responses
Tasks complete at faster speed, which you'll notice in the chat app, in streaming API responses and in long agent sessions.
Web search - fetch
The API can now search the web and fetch a page before answering, so your app can respond with current information and sources. Details in the search - fetch section.
Who benefits
| If you… | What changes for you | Read next |
|---|---|---|
| Use the chat app | Default answers come back faster, and bigger requests, such as multi-part plans and longer code, work better. | How to use DeepSeek |
| Build with the API | Lower cost per task, faster replies, and the option to add search - fetch. | API guide |
| Code with Harness | Flash sessions handle multi-file features and fixes more reliably and cheaply. | Harness Coding |
| Automate with Harness | Scheduled jobs and plugin workflows on Flash cost less and finish sooner. | Scheduled tasks · Plugins |
| Work with documents - spreadsheets | Multi-step jobs across many files complete in fewer rounds. | Documents · Spreadsheets |
| Pay for V4 Pro today | More of your workload may now run well on Flash, for less. | V4 Pro vs V4.1 Flash |
Do I need to change anything?
| Situation | Action |
|---|---|
| Chat app on the web | Nothing. You're using the new build already. |
| Mobile app | Keep the app updated from the official store links on our download page. |
| Existing API integration on Flash | Usually nothing. Keep the same Flash model name. Re-run your key prompts in case output style changed slightly. Confirm in DeepSeek's release notes whether the model name or behaviour changed. |
| Want web search - fetch | Optional: add the tools to your requests. See search - fetch. |
| Harness Desktop | Update Harness if prompted. Flash sessions use the new build. |
| Budgets - alerts | Review your spending after a week. Cost per task should go down, but search - fetch adds input tokens if you turn it on. |
How to try V4.1-Flash-1017
Just ask
Open the chat app and give it a multi-part task, such as planning a project with steps, owners and risks. Flash is the default.
Open chat → APISend a request
Call the Flash model as usual and compare speed and token use with your previous runs.
API guide → HarnessRun a project task
Ask Harness to add a small feature across several files and run the tests.
Harness guide →Add a CSV export button to the reports page, including the route, the template change and a test. Show the plan first, then make the changes and run the tests.The new search - fetch tools
Until this release, the API could only answer from what the model learned in training. Now a request can ask the model to search the web for relevant pages and fetch a page's full text before answering. Your app can handle questions about current events, prices, documentation and anything else that changes, and can show the sources it used.
- Opt-in per request, so normal requests are unchanged.
- Results add input tokens, so use them only when you need current information.
- Treat fetched content as data, not instructions. Learn about prompt injection.
How 1017 changes which model you should use
| Task | Before 1017 | After 1017 |
|---|---|---|
| Everyday chat, summaries, drafts | Flash | Flash, now faster |
| Multi-file coding - agent work | Often Pro | Flash for most tasks. Pro for architecture |
| Questions needing current info | Not possible in the API | Flash + search - fetch |
| Long, high-stakes documents | Pro | Pro, still the better choice |
| Maths, logic, hard bugs | R1 | R1, unchanged |
| IDE autocomplete | Coder | Coder, unchanged |
Re-test checklist for developers
- Run your saved test prompts on the new build and compare the outputs with your previous results.
- Check output formats. If you parse JSON or a fixed structure, make sure it still matches.
- Compare token usage from the
usagefield. You should see fewer tokens per task. - Measure response time, especially for streaming chat interfaces.
- Move traffic down from Pro where Flash now matches your quality bar.
- Pilot search - fetch on the questions that currently go stale or out of date.
V4.1-Flash-1017 - frequently asked questions
What is V4.1-Flash-1017?
An updated build of DeepSeek V4.1 Flash. It significantly improves project development, completes tasks at lower cost and faster speed, and comes with web search and fetch support in the API.
Do I need to change my code?
For most integrations, no. Keep using the Flash model as before. Re-test important prompts, and confirm any model-name details in DeepSeek's official release notes.
Is it free in the chat app?
Yes. The chat app on the web and the mobile apps is free, and it uses Flash by default.
How much cheaper is it?
DeepSeek says tasks are completed at lower cost. The saving depends on your workload. Compare the token counts in usage before and after, and see API pricing for current rates.
Does search - fetch cost extra?
Search results and fetched page text count as input tokens, and the built-in tools may have their own charge. Check the API platform and our search - fetch cost notes.
Is Flash now as good as V4 Pro?
It closes much of the gap for everyday and project work. V4 Pro still leads on long, nuanced, high-stakes tasks. See V4 Pro vs V4.1 Flash.
Is this DeepSeek's official announcement?
No. DeepSeekR1.com is an independent guide that explains releases in plain English. See about us, our disclaimer, privacy policy and terms.