DeepSeek Opens Its Search and Fetch APIs at $0.75 per 1,000 Calls

DeepSeek announced on X on October 17 that its Search and Fetch APIs are open to anyone with an API key, priced at $0.75 per 1,000 successful requests, with failed requests, empty results and timeouts not billed. Search returns 10 results by default and up to 50 with operators such as site:, lang:, after: and before:, while Fetch returns a URL's page as Markdown; each endpoint is capped at 5 requests per second per account. Server-side Search and Fetch calls through DeepSeek Harness, Claude Code and Codex are free for DeepSeek API users until billing starts at 16:00 UTC on October 31, 2026, and the APIs shipped alongside DeepSeek-V4.1-Flash-1017.

DeepSeek's model update got the headlines, but the part of Saturday's announcement that matters for developers is smaller and cheaper: its own web search and page-fetching tools, now open to anyone with an API key at $0.75 per 1,000 calls. - DeepSeek announced on X on October 17 that its Search and Fetch APIs are open to everyone, powered by what it calls its in-house search engine serving hundreds of millions of requests a day - Each endpoint costs $0.75 per 1,000 successful requests, and failed requests, empty results and timeouts are not billed - Server-side Search and Fetch calls made through DeepSeek Harness, Claude Code and Codex are free for DeepSeek API users until billing starts at 16:00 UTC on October 31, 2026 - Each endpoint is capped at 5 requests per second per account, which will matter for agents that fan out many searches at once - DeepSeek's privacy policy says it stores personal data in China, which matters if your queries are sensitive - It shipped alongside DeepSeek-V4.1-Flash-1017, an update DeepSeek says is faster and cheaper than the previous version What DeepSeek actually opened up In the third post of a four-part thread, DeepSeek wrote that its "Search & Fetch APIs are now open to everyone," powered by its in-house search engine, "serving hundreds of millions of requests daily worldwide." Each endpoint costs $0.75 per 1,000 calls. Search takes a query and returns ranked results, and Fetch takes a URL and returns the page as Markdown, which is the form language models handle best. The documentation fills in the details. Search accepts queries of 1 to 128 characters, returns 10 results by default and up to 50, and supports operators such as site: , lang: , after: and before: . Each result carries a title, URL, snippet, a longer summary meant for models, a relevance score from 0 to 3 and an authority score from 0 to 4. Page content is returned only when you ask for it. Both endpoints use the same bearer authentication as DeepSeek's chat API, so there is no separate key to set up. How the pricing works The $0.75 figure is per successful request, and the rules are developer-friendly. You are billed only when a search returns results or a fetch returns page content. Failed requests, empty results and timeouts cost nothing, and asking for 50 results costs the same as asking for one. One developer replying to the thread pointed out that this makes a large search with full page content the same price as a single hit. For context, one comparison published on launch day lists other search APIs per 1,000 calls at about $5 for Brave, $4 to $7 for Exa, $1 to $5 for Perplexity and $8 to $16 for Tavily. Those numbers come from a single outlet's snapshot, and the services are not identical, so treat them as a rough guide. The point stands that $0.75 is far below the usual range. A launch price can also change, and DeepSeek raised prices on another product earlier this year. Anthropic Turns Claude Code Into a Multi-Agent Team With New Projects Beta https://startupfortune.com/anthropic-turns-claude-code-into-a-multi-agent-team-with-new-projects-beta/ Anthropic relaunched Projects inside Claude Code on September 17, adding a coordinating agent that splits engineering goals across multiple parallel cloud sessions called threads. Each thread works its own branch and shares memory with the rest of the project, though execution is cloud-only for now and the beta is limited to select Pro and Max... - how to use Claude Code multi agent projects https://startupfortune.com/anthropic-turns-claude-code-into-a-multi-agent-team-with-new-projects-beta/ - Claude Code parallel cloud agents for engineering tasks https://startupfortune.com/anthropic-turns-claude-code-into-a-multi-agent-team-with-new-projects-beta/ Free through Claude Code and Codex, for now The detail that drew the most attention is that "Search & Fetch calls via server-side tools in DeepSeek Harness, Claude Code, and Codex are currently free for DeepSeek API users." Billing starts at 16:00 UTC on October 31, 2026, which DeepSeek's documentation gives as 00:00 Beijing time on November 1. It helps to be precise about what that means. The calls are free for people who point those tools at DeepSeek's API and use its search through them. It is not a free search feature for Claude Code or Codex users in general. DeepSeek is using the free window to get its search into other companies' coding agents, which fits a release it says is optimised for DeepSeek Harness and Tencent CodeBuddy. The limits that matter for agents Each endpoint is capped at 5 requests per second per account, and Fetch also has a per-domain limit that DeepSeek does not disclose. Search and Fetch have separate quotas, and going over returns an HTTP 429 error. Replies from developers raised the obvious worry: an agent that fans out many searches at once will hit that ceiling quickly, and empty results do not bill but still use up a slot. Others asked whether the 429 response says which limit was reached and when to retry, which the documentation does not spell out. Fetch can also fail in ways that are easy to misread. A 404 can mean the page could not be read, not that it is gone, and a 403 means DeepSeek's security policy rejected the request. One developer suggested treating these as retrieval failures in your own wrapper so an agent does not conclude that a source has been deleted. The data question The cheap price comes with a trade-off worth weighing. Queries and fetched URLs go to DeepSeek's hosted service, and its privacy policy says it "collects, processes and stores personal data in the People's Republic of China." Its platform terms are governed by Chinese law. The policy lists user input among the data used to improve its technology, and the API terms do not say clearly whether API inputs are used for training. The search guide says nothing about retention or region. If your queries include customer data or anything confidential, read the terms first, because regulators in Italy and Germany have acted against DeepSeek over data transfers to China. The model update behind it The search news came with DeepSeek-V4.1-Flash-1017, which DeepSeek says delivers stronger benchmark performance, faster task completion and shorter outputs than the previous version, "a Pareto improvement in performance and cost." It is live on DeepSeek's website, app and API, and the company says it uses reinforcement learning with self-feedback to build whole projects, from games to 3D models. The weights are on Hugging Face and ModelScope. DeepSeek's model card reports 77.0 on a coding benchmark called DeepSWE against 74.0 for Anthropic's Opus 5, but those are the vendor's own numbers on a table that leaves out newer rivals, and independent testing will say more. Reported parameter counts also differ between trackers, so we have left them out. What to take from it For anyone building agents, the practical read is simple. DeepSeek's search is cheap and, through the agents named above, free until the end of the month, so it is worth testing now. Check the rate limit against how your agent actually works, keep sensitive queries away from it until you have read the data terms, and expect the price to be reviewed once the free period ends. DeepSeek Ties Its New V4.1-Flash-1017 Model Tighter to Its Own Harness https://startupfortune.com/deepseek-ties-its-new-v41-flash-1017-model-tighter-to-its-own-harness/ DeepSeek told users to upgrade to DeepSeek Harness v0.2.1-rc.1 for the best experience with the new model, which posted a 74.2% pass rate on Deep-SWE v1.1, edging out Opus-5's 74.0%. Tencent's CodeBuddy and WorkBuddy both get optimized access in the same announcement. - DeepSeek V4.1 Flash model performance benchmark improvements https://startupfortune.com/deepseek-ties-its-new-v41-flash-1017-model-tighter-to-its-own-harness/ - DeepSeek new API model faster cheaper October release https://startupfortune.com/deepseek-ties-its-new-v41-flash-1017-model-tighter-to-its-own-harness/ This article is posted in AI News https://startupfortune.com/category/ai/ , check it out for more related stories. Join the discussion Open in the community → https://startupfortune.com/community/ Almost there. Sign in and your reply posts straight away.