The Pulse
DeepSeek Opens V4.1 Flash Weights and Search APIs
DeepSeek released the open-weight V4.1-Flash-1017 model and opened its Search and Fetch APIs on October 17, 2026. The company lists a 77.0 score on DeepSWE v1.1 and charges $0.75 per 1,000 successful requests to either API endpoint.

DeepSWE score rises to 77.0
DeepSeek-V4.1-Flash-1017 scores 77.0 on DeepSWE v1.1, up from 74.2 for the previous V4.1-Flash release. DeepSeek announced the model on October 17, 2026, alongside newly public Search and Fetch APIs. The update, the company says, completes tasks faster and produces shorter outputs, which can reduce the cost of running agent tasks.
DeepSeek describes the revision as a “Pareto improvement in performance and cost.” The company’s change log also reports gains on Terminal-Bench 4.0, from 31.2 to 42.3, and Automation-Bench v1.0.6, from 54.8 to 60.5. Terminal-Bench used the Claude Code framework and DeepSWE used the Mini-SWE-V2 framework, DeepSeek says, so the scores reflect those stated test setups.
“Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost.”
DeepSeek’s announcement
Weights carry an MIT license
The new checkpoint is live on DeepSeek’s website, app and API, and DeepSeek lists downloads through Hugging Face and ModelScope. The Hugging Face model card identifies the repository and weights as MIT-licensed and includes instructions for running the model with tools such as Transformers, vLLM and SGLang. That gives developers a route to use the released weights outside DeepSeek’s hosted service, subject to the computing resources needed to run them.
The company recommends DeepSeek Harness version 0.2.1-rc.1 or later for the model. It also names Tencent CodeBuddy and WorkBuddy among the platforms it optimized for. The announcement does not give new API model pricing; DeepSeek’s change log says the API calling method and price remain unchanged, with the model available under the existing deepseek-flash identifier.
Search and Fetch add live web access
The new Search API returns web pages for a query, while Fetch retrieves a specified page and returns its content as Markdown. Both use JSON requests and the same bearer-key authentication as DeepSeek’s model API. Search requests can return up to 50 results; Fetch handles one URL per request.
DeepSeek’s search engine processes hundreds of millions of requests daily and indexes hundreds of billions of pages, the company says. The search engine also filters SEO spam, GEO manipulation and low-quality AI-generated content, according to the company. Those are DeepSeek’s descriptions of its own search system, not independently reported measurements of result quality.
Each successful request costs 0.075 cents
DeepSeek prices Search and Fetch separately at $0.75 per 1,000 successful requests. A request is billable only when it returns a result; an empty search or failed fetch is not charged, and asking Search for more results does not change the per-request fee. Each endpoint has an account-level limit of five requests per second.
For API users accessing the tools through server-side features in DeepSeek Harness, Claude Code or Codex, the announcement says Search and Fetch calls are free until October 31, 2026, at 16:00 UTC, when billing begins. The change gives developers a single provider for an open-weight agent model and the web retrieval tools it can call, while making the ongoing cost depend on how often those tools run.
Sources
- Original source DeepSeek API Docs · page 1 api-docs.deepseek.com
Other pages read
- DeepSeek API Docs · page 2 api-docs.deepseek.com
- DeepSeek API Docs · page 3 api-docs.deepseek.com
- Hugging Face huggingface.co