DeepSeek upgraded its V4.1 Flash model to a new version, DeepSeek-V4.1-Flash-1017, on Oct. 17, according to the company’s API change log. The same day, it opened search and fetch APIs powered by its own search engine to all developers.
Nothing changes in how developers call the model. DeepSeek says the API name, deepseek-flash, and the pricing are the same, and its web and mobile apps also switched to the new version. The company says it cut the model’s task runtime and output length, so tasks cost less and finish faster.
DeepSeek says it tuned the update for its own DeepSeek Harness and for Tencent’s CodeBuddy and WorkBuddy tools, among others, and recommends DeepSeek Harness version 0.2.1-rc.1 or later. In an X post quoted by PPC Land, the company said the model “uses reinforcement learning with self-feedback to improve end-to-end project development.”
How it scores
DeepSeek reports 77.0 on DeepSWE v1.1, a software engineering benchmark, up from 74.2 for the original V4.1 Flash in September. It reports 42.3 on Terminal-Bench 4.0, which tests command-line tasks, up from 31.2, and 35.4 on Agents’ Last Exam, a test of professional work, up from 31.8. These are the company’s own results.
DeepSeek says Terminal-Bench was run using Anthropic’s Claude Code as the agent framework, and DeepSWE with the Mini-SWE-V2 framework. The change log lists other gains, including 60.5 on Automation-Bench and 68.2 on NL2Repo-Bench, a test of building a code repository from a written description.
The search API
DeepSeek says its search engine indexes hundreds of billions of web pages and handles hundreds of millions of requests a day. It says the engine filters out search spam and low-quality AI-generated content.
There are two endpoints. Search “finds web pages for a query and returns structured results,” and Fetch “retrieves a single web page and returns its content as Markdown,” according to DeepSeek’s guide as quoted by PPC Land. Both use the same API key as DeepSeek’s models.
Each costs $0.75 per 1,000 calls, according to PPC Land. Server-side web lookups made through Claude Code, Codex and DeepSeek Harness when they use DeepSeek models are free until 16:00 UTC on Oct. 31, which is noon Eastern, the outlet reported.
DeepSeek’s API reference shows what comes back. Each result includes a short snippet meant for display, a longer summary written for a model to read, a relevance score from 0 to 3 and an “authority tier” from 0 to 4. Developers can use operators such as site: and quoted phrases to narrow a query.
PPC Land flagged two optional fields. One passes along the user’s original question and the other takes an IP address to localize results. Neither documentation page says how DeepSeek stores or uses those values, the outlet reported.
It noted that Italy’s data protection authority banned DeepSeek’s consumer app in January 2025, in part over transfers of personal data to China.
The outlet also noted that DeepSeek doesn’t explain how a site earns its authority tier.
Source: DeepSeek API Docs (change log), DeepSeek, “DeepSeek API change log: DeepSeek-V4.1-Flash-1017”
Additional sources: PPC Land, “DeepSeek prices web search for AI agents at $0.75 per 1,000 calls”; DeepSeek API Docs, DeepSeek, “Search”
