DeepSeek V4.1 Flash 1017: What Changed and How to Use It Free
DeepSeek upgraded V4.1 Flash to the DeepSeek-V4.1-Flash-1017 build on October 17, 2026. The API name and price did not change — calls still use deepseek-flash — and DeepSeek's own web and app were updated to the same version the same day. The official chat is still free but still needs an account; what changed is the model behind the name, not the way you get in. aiwith's DeepSeek Flash is served through that same deepseek-flash name, so the version you get here is the 1017 build.
Last checked: October 17, 2026. This is a source-checked release guide, not a hands-on benchmark.
What is DeepSeek-V4.1-Flash-1017?
It is a same-generation update to V4.1 Flash, not a new model family. DeepSeek published it in its API changelog under October 17, 2026, with three points that matter:
- The API is unchanged. "API 调用方式与价格保持不变" — the calling convention and the price stay the same, and
deepseek-flashstill returns the latest model. - The web and app were updated too. The free official chat at chat.deepseek.com now serves the 1017 build.
- The "-1017" is a build tag (October 17), the same way earlier DeepSeek releases were named.
DeepSeek frames the release as a "performance–cost Pareto improvement": higher scores and shorter task run time and output length, so a task finishes faster and generates fewer tokens.
What changed in the 1017 build?
DeepSeek reports broad gains over the previous V4.1 Flash across agentic, coding and reasoning benchmarks. The numbers below are vendor-reported and pulled from the two official changelog entries (V4.1 Flash, September 10, 2026 → V4.1-Flash-1017, October 17, 2026):
| Benchmark | V4.1 Flash (Sept 2026) | V4.1-Flash-1017 |
|---|---|---|
| HLE (w/o tools) | 36.8 | 42.6 |
| HLE (w/o tools, text-only subset) | 39.1 | 44.0 |
| HLE (w/tools) | 63.9 | not listed |
| Terminal-Bench 4.0 | 31.2 | 42.3 |
| DeepSWE v1.1 | 74.2 | 77.0 |
| ProgramBench | 20.3 | 29.1 |
| NL2Repo-Bench | 65.4 | 68.2 |
| Automation-Bench | 54.8 | 60.5 |
| Agents' Last Exam | 31.8 | 35.4 |
| Chartography (w/tools) | 78.9 | 83.0 |
| BabyVision (w/tools) | 89.6 | 92.5 |
| ZeroBench-main (w/tools) | 49.0 | 51.0 |
Two caveats on reading the table. The September entry lists an HLE figure with tools (63.9) that the 1017 entry does not repeat, so there is no like-for-like "w/tools" comparison; and DeepSeek marks the 1017 HLE numbers as (w/o tools), with the 44.0 figure taken on the text-only subset rather than all modalities.
Two other notes from the changelog: Terminal-Bench 4.0 was scored with the Claude Code harness and DeepSWE with Mini-SWE-V2, and ZeroBench-main used Pass@5. DeepSeek also says it tuned the model for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy, and recommends DeepSeek Harness 0.2.1-rc.1 or newer. If you run DeepSeek through a coding harness, upgrading it is part of getting the new behavior.
Is it still free, and did the price change?
Two separate questions:
- DeepSeek's API price did not change. The changelog says so directly. V4.1 Flash has been priced below the frontier proprietary models since launch, and that stays true for 1017.
- The official chat is still free after signup. There is no paid tier that unlocks the 1017 build; it is just what chat.deepseek.com serves now. The one thing it still requires is an account.
On aiwith, DeepSeek has always been the value option, and the model key is unchanged — this is a version bump behind the same name, not a new line item.
How do I use the 1017 build without an account?
The access rules did not change with this release. chat.deepseek.com still asks for phone or email verification before your first message, and there is still no guest mode and no "try without signing in" button. If avoiding a new account is the real goal, the three routes are:
- Google sign-in on chat.deepseek.com — an account, but not a new one.
- A third-party front-end with guest access — genuinely account-free, but rate-limited, no history, no uploads.
- A multi-model platform — one signup instead of one per model.
We break down which of these are real, and their limits, in DeepSeek Without Login. For the "is DeepSeek free?" framing, see Is DeepSeek Free?.
Does aiwith serve the 1017 build?
Yes. aiwith calls DeepSeek under its official deepseek-flash model name, and DeepSeek routes that name to the latest V4.1 Flash build — the 1017 release. So the "DeepSeek V4.1 Flash" you select on aiwith is the new version, with the same pricing and the same API.
New aiwith accounts start with free trial credits and no credit card, and DeepSeek sits alongside GPT-6, Claude, Gemini and Kimi in the same chat — useful when you want to sanity-check a DeepSeek answer against another model without a second signup. Paid plans start at $9.9/month. See DeepSeek on aiwith for the current model list and pricing.
DeepSeek also opened a search API
Separately on the same changelog date, DeepSeek opened its self-developed search API, powered by an AI-native engine it says handles hundreds of millions of queries a day and indexes hundreds of billions of web pages across languages (see the search guide). It is a building-block release rather than a chat-model change — relevant if you are adding web search to a product, not if you just want to chat with the 1017 model.