DeepSeek updates its low-cost AI model and opens its own search engine to developers

The Chinese AI company DeepSeek released an updated version of its Flash model, called DeepSeek-V4.1-Flash-1017, on October 17, 2026. At the same time, the company is opening its own search engine to developers for the first time.

Faster, with shorter answers

According to DeepSeek, the update delivers stronger test results, faster task completion and shorter answers. The model is served under the existing name deepseek-flash in the company's API. An API is a technical interface that lets other programs communicate with a service. Existing users therefore don't need to change their code. Older model names still work, but the underlying models have been retired.

On the DeepSWE v1.1 coding benchmark, the model scored 77.0 percent, according to DeepSeek's own figures. The outlet BigGo Finance reports that the score on that test rose from 74.2, and that the score on another test, Terminal-Bench 4.0, rose from 31.2 to 42.3. According to BigGo, pricing for the model is unchanged, and a task on DeepSWE costs about $0.19, compared with about $0.49 for the previous version.

BigGo also writes that the model outperforms several competitors, including Claude Opus 5 and GPT-5.6-Sol, in a number of areas. This is a restatement of DeepSeek's own results and has not been independently verified. No detailed technical paper on the model's architecture has been published either.

According to BigGo, the model is open source under the MIT license and is available on Hugging Face and ModelScope. Sources disagree on the model's size, so the parameter count should be treated with caution.

Own search engine opens up

With a new Search/Fetch API, developers can retrieve search results and web pages as text directly from DeepSeek's own search engine. This can be used, for example, to give an AI service access to fresh information from the web.

According to DeepSeek, as relayed by BigGo, the search engine indexes hundreds of billions of web pages, handles hundreds of millions of searches a day and filters out low-quality content.

According to BigGo, search and fetch calls cost 5 yuan, about $0.75, per 1,000 calls. Access is free through the end of October. According to Tech Insider, paid usage begins on October 31, 2026, at 4:00 p.m. UTC, while BigGo cites November 1. The difference may be due to time zones.

For ordinary users, the news mainly means that AI services built on DeepSeek could become cheaper and faster. Whether the model lives up to the numbers in practice will have to be shown by independent tests.