Michael JohnsonOn October 17, DeepSeek updated V4.1 Flash to version 1017, published the model weights, and opened web search and page retrieval APIs to developers. The changes affect two distinct layers: the model keeps its existing API identifier, while Search and Fetch are available as separate paid interfaces.
In its announcement, DeepSeek says the new version is already available on its website, in its app, and through its API. The model can still be called using the name deepseek-flash; the documentation says that neither the API method nor the model’s pricing has changed. The weights for DeepSeek-V4.1-Flash-1017 have been published on Hugging Face and ModelScope.
What changes for teams using the API
The main practical change for model integration is that the target version behind the existing deepseek-flash name has been updated. Applications can therefore keep sending requests using the existing schema, but teams should test the model on their own coding and agent workloads before deploying it widely. DeepSeek reports improved benchmark results, faster execution, and shorter responses; these are company-reported assessments, not an independent comparison.
To check compatibility, it is useful to rerun a set of typical prompts and record not only output quality, but also execution time, response length, and total cost. In its DeepSWE benchmark, the company compares task costs in the mini-swe-v2 environment; those results apply to the specified test scenario. If an application depends on exact model behavior, record results before and after the update and provide a way to roll back in your configuration.
Search and page retrieval have separate pricing
DeepSeek has opened Search API for web searches and Fetch API for retrieving page content. According to the announcement, each interface costs $0.75 per 1,000 calls. The company describes its own search index and says it processes hundreds of millions of queries a day; these figures come from the provider itself.
Server-side Search and Fetch calls through DeepSeek Harness, Claude Code, and Codex are priced separately. DeepSeek says they are currently free for API users, with charges set to begin on October 31, 2026, at 16:00 UTC. This is a concrete budget change for teams already using server-side search in agent workflows; direct API calls have the announced per-1,000-call price.
Before integrating the services, check which route your application uses: direct calls to Search or Fetch, or server-side tools in an agent environment. Then estimate the expected number of calls separately for search and page retrieval, and revisit the calculation by October 31. This distinction helps identify which costs apply immediately and which will start accruing later.
What the evidence says about performance
The release combines a model update, the launch of search interfaces, and publication of the weights, but the performance evidence in the announcement consists primarily of DeepSeek’s own tests. To decide whether to switch, developers should compare those claims against their product’s workloads: task success, latency, token usage, and the number of search calls.
The integration takeaway is simple: the existing model name remains, while web search and page retrieval require separate call tracking. Teams already using DeepSeek’s server-side tools should budget for a new expense starting October 31 and test how the updated version performs on their own production prompts.