Michael JohnsonOn October 17, DeepSeek released V4.1 Flash-1017 and said the updated model was already available on its website, in its app, and through its API. That same day, the company opened the Search & Fetch APIs to all users—two tools for web search and retrieving web pages. Developers should note three distinct changes: a new model version, access to its weights, and upcoming charges for search calls.
In its announcement, DeepSeek says the new version performs better on tests, completes tasks faster, and produces shorter responses than its predecessor. The company also describes training with reinforcement learning and self-checking to create entire projects. The published metrics and demonstrations should be understood as DeepSeek’s own claims: the announcement does not provide independent results.
What changes for developers
The V4.1 Flash-1017 weights are available on Hugging Face and ModelScope, allowing teams to examine the model files and test a suitable way to run them in their own environment. Before migrating, it is useful to record the current version, select a few representative tasks, and compare quality, execution time, and response length using the same inputs. This test will show how well the claims about shorter, faster responses apply to a particular workflow.
Web search and page-content retrieval are a separate part of the update. DeepSeek lists a price of $0.75 per 1,000 calls for each endpoint. Calls to Search & Fetch through DeepSeek Harness, Claude Code, and Codex server-side tools are currently free for DeepSeek API users; the company has set billing to begin on October 31, 2026, at 16:00 UTC. That is 12:00 p.m. Eastern Time.
How to estimate costs before switching
First, separate expected traffic by operation type: search and page retrieval are billed separately. Then estimate the number of calls to each tool over the period and multiply each figure by $0.75 per thousand. For example, 20,000 search calls and 8,000 fetch calls at the stated rate would cost $21 for that volume, excluding any other potential API charges.
Before October 31, check which stages of your application actually need web search, how often the same query is repeated, and whether every result needs to be fetched. If your infrastructure uses server-side tools, also verify the billing start date and how usage will be counted once charges begin: the free tier currently applies to the specified DeepSeek integrations and API users.
The update combines a new model release with web tools and the publication of model weights. A practical next step is to test the model on your own tasks, estimate Search and Fetch volumes separately, and add the billing start date to your budget and monitoring in advance.