DeepSeek opens its search engine to developers at $0.75 per 1,000 searches and updates V4.1 Flash

DeepSeek logo above the text Search API: $0.75 per 1,000, plus a V4.1 Flash update at the same price

Key points

  • Developers get access to DeepSeek's search engine
  • How DeepSeek's search price compares with rivals
  • Where DeepSeek's V4.1 Flash scores rank

DeepSeek opened its in-house search engine to outside developers on Saturday at a lower listed price than some rival search services. Developers pay $0.75 per 1,000 successful searches. Brave lists $5 per 1,000 requests for its search API, and Exa's search starts at $4, though the services' features differ.

DeepSeek also updated V4.1 Flash, saying the model completes tasks faster and scores higher on coding and agent benchmarks without an increase in API prices. The new version, DeepSeek-V4.1-Flash-1017, also produces shorter answers, the company said. Both releases are listed in its Oct. 17 API change log.

DeepSeek opens its search engine

DeepSeek's chatbot and API already used the company's own search engine, which DeepSeek said "reliably handles hundreds of millions of search requests from around the world every day" and "indexes hundreds of billions of web pages." The company said it filters out "SEO spam, GEO manipulation, and low-quality AIGC." AIGC stands for AI-generated content. GEO refers to generative engine optimization, efforts to influence how AI systems present information.

DeepSeek said the filtering is meant to provide "authoritative and trustworthy information for users and LLMs." Search APIs like these let AI models and agents pull live web results into their answers.

Outside developers can now use the engine directly through two APIs. DeepSeek charges $0.75 per 1,000 successful requests for Search and the same for Fetch, which pulls content from web pages, according to its pricing page. Brave lists its search API at $5 per 1,000 requests, and Exa's search starts at $4 per 1,000 requests.

What changed in V4.1 Flash

The update improves on the original Sept. 10 release across all six benchmarks shown below, according to DeepSeek's own testing. The largest score increases were on Terminal-Bench 4.0 and ProgramBench.

BenchmarkSept. 10 versionOct. 17 version
DeepSWE v1.174.277.0
Terminal-Bench 4.031.242.3
ProgramBench20.329.1
NL2Repo-Bench65.468.2
Agents' Last Exam31.835.4
Humanity's Last Exam, no tools36.842.6

DeepSeek said it tested Terminal-Bench 4.0 through Anthropic's Claude Code and DeepSWE through Mini-SWE-V2, a framework for running the tasks.

V4.1 Flash is the smallest model in DeepSeek's new architecture family, according to the company. DeepSeek said it optimized the update for its own DeepSeek Harness, Tencent's CodeBuddy, and WorkBuddy.

The model's weights are available on Hugging Face under the MIT license, which permits commercial use.

How the scores compare

On DeepSWE v1.1, DeepSeek's 77.0 would trail only Google's Gemini 4 Argon, at 77.9%, among scores on the benchmark's public tracker. OpenAI reported 75.2% for GPT-6.1 Sol, and Anthropic reported 74.2% for Claude Opus 5.5. Those scores, like DeepSeek's, are the companies' own results, and the labs ran the tasks in different setups.

On Terminal-Bench 4.0, DeepSeek's 42.3% trails the leaders on the benchmark's public leaderboard. Claude Opus 5.5 scored 64.8% running in Claude Code, the same tool DeepSeek used, and OpenAI's GPT-6 Astra and GPT-6.1 Sol each scored 58.2% in Codex. Z.ai's GLM-5.3 scored 41.8% in Claude Code. DeepSeek's new result isn't listed on that leaderboard.

V4.1 Flash's API prices stay unchanged

During peak hours, DeepSeek charges $0.30 per million uncached input tokens and $1.20 per million output tokens. Off-peak rates are half that, and peak hours run from 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays.

Developers who still call the older V4 Flash model names are now routed to the new version and billed at the Flash price.

The release comes as DeepSeek raises money. DeepSeek was close to raising at least 80 billion yuan, about $11.9 billion, in early October, with Tencent and CATL among its biggest backers. The Standard reported on Oct. 13 that the round had reached 100 billion yuan. DeepSeek hasn't confirmed the figures.

Frequently asked questions

What is DeepSeek-V4.1-Flash-1017?

DeepSeek-V4.1-Flash-1017 is an updated version of DeepSeek's V4.1 Flash model, released October 17, 2026. DeepSeek says it scores higher on its coding and agent benchmarks than the September 10 version, finishes tasks faster, and gives shorter answers. Its weights are posted on Hugging Face under the MIT license.

How much does DeepSeek V4.1 Flash cost?

Pricing did not change with the October 17 update. During peak hours, DeepSeek charges $0.30 per million uncached input tokens and $1.20 per million output tokens through its API, and off-peak rates are half that.

What is DeepSeek's Search API?

On October 17, 2026, DeepSeek opened its in-house search engine to outside developers through new Search and Fetch APIs. Each costs $0.75 per 1,000 successful requests. DeepSeek says the engine indexes hundreds of billions of web pages.

How does DeepSeek's search API price compare with Brave and Exa?

DeepSeek charges $0.75 per 1,000 successful search requests. Brave lists its search API at $5 per 1,000 requests, and Exa's search starts at $4 per 1,000 requests, according to their pricing pages as of October 17, 2026. These are listed prices for services whose features differ, so they aren't a like-for-like comparison.

How does DeepSeek V4.1 Flash compare with Claude and GPT on coding benchmarks?

DeepSeek reported 77.0% on DeepSWE v1.1 for V4.1-Flash-1017, behind only Google's self-reported 77.9% for Gemini 4 Argon on the benchmark's public tracker. On Terminal-Bench 4.0, DeepSeek reported 42.3%, well behind Claude Opus 5.5 at 64.8% on that benchmark's public leaderboard. DeepSeek's scores are its own results.

More on DeepSeek

Latest stories

Dennis Singleton
Dennis Singleton

Dennis Singleton was born in Australia and later moved to the United States. He has spent years following the markets, but what keeps his attention is how AI is built. He writes about the companies behind the technology, from semiconductor designers and advanced packaging to photonics, memory, networking, and the hardware powering modern AI. His approach starts with filings, earnings, and industry research, then translates the important details into clear, straightforward analysis without unnecessary hype.