DeepSeek upgrades its Flash model with V4.1-Flash-1017
On 17 October 2026, DeepSeek released DeepSeek-V4.1-Flash-1017, a minor update to its V4.1-Flash model from September. DeepSeek says it scores higher on reasoning, coding and agent tests and uses shorter answers. Its web chat and app already run it, API pricing is unchanged, and the weights are open under the MIT licence.
The short version
- 01DeepSeek released V4.1-Flash-1017 on 17 October 2026, a minor update to V4.1-Flash.
- 02On Terminal-Bench 4.0 it scores 42.3, up from 31.2, DeepSeek says.
- 03DeepSeek says tasks now take less time and use shorter answers.
- 04API pricing is unchanged, and the weights are under the MIT licence.
- 05The scores are DeepSeek's own and have not been independently tested.
At a glance
| Released | 17 October 2026 |
|---|---|
| Model | DeepSeek-V4.1-Flash-1017 |
| Replaces | DeepSeek-V4.1-Flash (10 September 2026) |
| API name | deepseek-flash, pricing unchanged |
| Licence | MIT |
| Context window | 1M tokens |
| Status | Confirmed |
What is DeepSeek-V4.1-Flash-1017?
It is a minor update to DeepSeek-V4.1-Flash, according to its model card on Hugging Face. DeepSeek says the original V4.1-Flash is the smallest model in its new architecture family and can understand images natively.
The original came out on 10 September 2026.
How much better is it?
DeepSeek's own scores show gains. On HLE without tools it moves from 36.8 to 42.6. On Terminal-Bench 4.0, a coding test run with the Claude Code harness, it moves from 31.2 to 42.3. On DeepSWE v1.1 it moves from 74.2 to 77.0.
In DeepSeek's own table, Anthropic's Opus 5 still scores higher on Terminal-Bench 4.0 (51.8) and HLE (56.3). These are the company's numbers, not independent tests.
Is it faster and cheaper?
DeepSeek says the model cuts task run time and output length, so users finish tasks faster and at lower cost. The API price itself is unchanged.
Who can use it?
DeepSeek's official web chat and app already use the new version. Developers reach it through the existing name, deepseek-flash. The weights are licensed under the MIT licence, and the card recommends a context window of 1M tokens.
Why does it matter?
An open model keeps improving without a price rise. If the scores hold up, people can run a stronger model at the same cost.
What's next?
Independent tests of the new scores have not been published yet. DeepSeek recommends upgrading to DeepSeek Harness 0.2.1-rc.1 or later for the best results.
Questions people ask
What is DeepSeek-V4.1-Flash-1017?
A minor update to DeepSeek's V4.1-Flash model, released on 17 October 2026.
Did the price change?
No. DeepSeek says the API calling method and pricing remain unchanged.
How do developers get it?
Use the model name deepseek-flash in the API.
Is it open source?
The weights are released under the MIT licence on Hugging Face.
Are the scores independent?
No. They are DeepSeek's own.
Sources
- PrimaryDeepSeek API Docs: Change Log (17 Oct 2026)
- PrimaryDeepSeek on Hugging Face: DeepSeek-V4.1-Flash-1017
Why "Confirmed"? We label a story "Confirmed" when the company or government announced it itself. Spotted an error? Email hello@openloopnews.com and we will correct it.
