DeepSeek says DeepSeek-V4.1-Flash-1017 is live on its website, app and API, and that it has opened Search and Fetch to developers. The October 17, 2026 announcement pairs the model update with two web-retrieval tools, each priced at $0.75 per 1,000 successful requests. DeepSeek’s announcement DeepSeek’s Search and Fetch guide

Search finds results; Fetch retrieves a page

Search returns structured web results for a query. Fetch retrieves one specified page and returns its extracted content as Markdown. Both use JSON requests and DeepSeek API bearer authentication.

APIWhat it returnsPrice per 1,000 successful requestsWhen a request is billableAccount limit
SearchStructured web results$0.75HTTP 200 with a non-empty results array5 requests per second
FetchOne page’s extracted Markdown$0.75HTTP 200 with page content5 requests per second, plus a per-domain limit

The listed rates are in U.S. dollars. DeepSeek does not charge for failed or timed-out requests, and an empty Search result is not billable. Search can return between 1 and 50 results, with 10 as the default; changing that count does not change the price of a successful request. The two endpoints have separate account-level limits, while Fetch also has a per-domain limit. DeepSeek’s Search and Fetch guide

When paid billing is scheduled to begin

DeepSeek says server-side Search and Fetch calls through DeepSeek Harness, Claude Code and Codex are free for DeepSeek API users at the time of the announcement. The company scheduled the start of per-request billing for those calls for October 31, 2026, at 16:00 UTC. DeepSeek’s announcement

These retrieval charges are separate from model token charges. NeoTeo’s earlier look at DeepSeek V4.1 Flash API pricing explains how cache reuse affects that separate meter.

DeepSeek’s reported benchmark result

MarktechPost reported a 77.0 Pass@1 score for DeepSeek-V4.1-Flash-1017 on DeepSWE v1.1, attributing the result to DeepSeek. DeepSeek’s release notice also links a benchmark chart. MarktechPost’s report