The DeepSeek V4.1 Flash 1017 update
DeepSeek-V4.1-Flash-1017 is the update DeepSeek shipped on 17 October 2026. It is the same model family with better agent and coding scores and shorter answers, served under the same API name at the same price.
What 1017 is
DeepSeek's changelog entry for 17 October 2026 says DeepSeek V4.1 Flash "has been upgraded to DeepSeek-V4.1-Flash-1017", and that the official web and app models moved to it the same day. The model card calls it "a minor update to DeepSeek-V4.1-Flash". The number is a date, the same convention as V4 Flash 0731.
On DeepSeek's own API nothing changes for the caller: the model name deepseek-flash now reaches 1017, and the changelog says "the API calling method and pricing remain unchanged". The weights are published on Hugging Face under the MIT licence.
DeepSeek's benchmark table: 1017 against V4.1 Flash
| Benchmark | 1017 | V4.1 Flash (September) |
|---|---|---|
| HLE, no tools | 42.6 | 36.8 |
| Terminal-Bench 4.0 | 42.3 | 31.2 |
| DeepSWE v1.1 | 77.0 | 74.2 |
| ProgramBench | 29.1 | 20.3 |
| NL2Repo-Bench | 68.2 | 65.4 |
| Automation-Bench v1.0.6 | 60.5 | 54.8 |
| Agents' Last Exam | 35.4 | 31.8 |
| Apex Bench v1.1 | 72.1 | 64.6 |
| Chartography (with tools) | 83.0 | 78.9 |
| BabyVision (with tools) | 92.5 | 89.6 |
| ZeroBench-main (with tools) | 51.0 | 49.0 |
From the DeepSeek-V4.1-Flash-1017 model card. DeepSeek ran HLE on its text-only subset, Terminal-Bench 4.0 in the Claude Code framework, DeepSWE with Mini-SWE-V2 and ZeroBench-main at pass@5.
The gains are largest on the agent benchmarks: Terminal-Bench 4.0 rose by about a third and ProgramBench by more than 40 percent. DeepSeek also says the update finishes tasks faster and writes shorter output, so a solved task costs less even at the same per-token price, and it recommends DeepSeek Harness 0.2.1-rc.1 or later to get the most from it.
What did not change
- Context. Still a million tokens. The card recommends a max_tokens of at least 256K, temperature 1.0 and top_p 1.0.
- Images. Image input is not new in 1017. V4.1 Flash shipped on 10 September 2026 with native visual understanding (DeepSeek's announcement); 1017 improves the vision scores rather than adding vision.
- The architecture. V4.1 Flash is a 552B mixture-of-experts model with 8B parameters active per token in prefill and 16B in decode, per the V4.1 Flash card. 1017 is an update to those weights, not a new size.
The same changelog entry also opened DeepSeek's Search and Fetch API, which is a separate product from the model.
1017 and the uncensored build
Wild West API sells DeepSeek V4.1 Flash Outlaw (deepseek-v4.1-flash-outlaw), an uncensored build of DeepSeek V4.1 Flash. It is text only, it calls tools and it reasons by default. We do not claim the 1017 improvements for it: the table above is DeepSeek's measurement of DeepSeek's checkpoint, and an uncensored build is a different set of weights that nobody benchmarks separately.
What carries over is the shape of the model: the million-token context, tool calling and the reasoning modes. Artificial Analysis scores the base V4.1 Flash at 39 on its Intelligence Index, which places it between GLM 5.3 Flash (42) and MiMo V2.6 Flash on the line.
FAQ
What is DeepSeek V4.1 Flash 1017?
The 17 October 2026 update to DeepSeek V4.1 Flash. DeepSeek describes it as a minor update with better reasoning, coding, agent and vision scores and shorter outputs. The number is the release date.
Did the price change with 1017?
Not on DeepSeek's own API. Its changelog says the API calling method and pricing are unchanged, and deepseek-flash now points at 1017.
Does 1017 add image input?
No. V4.1 Flash already had native visual understanding when it launched on 10 September 2026. 1017 raised its vision benchmark scores.
Is there an uncensored DeepSeek V4.1 Flash 1017?
Wild West API sells DeepSeek V4.1 Flash Outlaw, an uncensored build of DeepSeek V4.1 Flash, as deepseek-v4.1-flash-outlaw. It is text only, and we do not claim DeepSeek's 1017 benchmark gains for it.
Where are the 1017 weights?
On Hugging Face as deepseek-ai/DeepSeek-V4.1-Flash-1017, under the MIT licence.