DeepSeek updates V4.1-Flash: open weights under the MIT license and the same API price
Redacción IvokaOct 18, 9:00 AM 3 min
On October 17, DeepSeek released V4.1-Flash-1017, available on its web, app and API. It reports shorter answers and, according to one report, is opening its search and page-fetch APIs to outside developers.
On October 17, DeepSeek released DeepSeek-V4.1-Flash-1017, an update to the V4.1-Flash it launched in September. Its own Hugging Face model card calls it a minor update.
What changes
- Available now: DeepSeek says the model is available on its website, mobile app and developer platform, and that its official web and app models now run the new version. The API is reached under the same model name as before (
deepseek-flash). - Same price and calling method: DeepSeek says the API calling method and pricing are unchanged.
- Open weights: the weights were released under the MIT license on Hugging Face and ModelScope. Before using it in a commercial product, check the license file on the model page.
- Shorter answers and faster tasks: DeepSeek says the update reduced task runtime and output length. One report notes that since usage is billed per generated text, shorter outputs should lower the bill for anyone paying per use.
- Search and Fetch: according to one report, DeepSeek is opening its search and page-fetch APIs to outside developers for the first time. DeepSeek says they are currently free when reached through server-side tool calls in DeepSeek Harness, Claude Code or Codex, with billing ($0.75 per 1,000 successful requests) starting on November 1, 2026.
A caution on the numbers
The scores DeepSeek publishes (for example, 42.3 on Terminal-Bench 4.0 versus 31.2 for the September model) are figures reported by the company itself, and the sources do not show them being independently replicated. They help compare against its previous version, but they do not replace testing it on your own tasks.
What to do with this
If you do not code, DeepSeek says its web and app already use the new model, but the announcement does not spell out whether it replaces the old one by default or how it rolls out by country; check the model label in the app before assuming you are on the new version. If someone builds tools for your business, they can test this model as a low-cost, open-weights option.