DeepSeek Search API and Fetch API: $0.75 per 1,000 Successful Requests

DeepSeek has added a Search API and a Fetch API so agents and RAG applications can pull live web content. Search returns web results, and Fetch returns the body of a single page as Markdown. Both cost $0.75 per 1,000 successful requests, and a request only counts as successful if it returns HTTP 200 with an actual result.

DeepSeek has added a Search API and a Fetch API so agents and RAG applications can pull live web content. Search returns web results, and Fetch returns the body of a single page as Markdown. Both cost $0.75 per 1,000 successful requests, and a request only counts as successful if it returns HTTP 200 with an actual result.

What each endpoint does

The two APIs split a common agent workflow into two steps. Search finds candidate pages for a query. Fetch takes one URL and returns the page's main content in Markdown.

That matters because language models have a training cutoff. Anything that happened after it, such as a new library release, a price change or a news item, has to be supplied at inference time. Search and fetch are the usual way to do that.

The billing rule is the interesting part

The price is identical for both endpoints: $0.75 per 1,000 successful requests, or $0.00075 per call. Billing applies only when the response is HTTP 200 and contains real results.

This counts more than it first appears. Agents that browse the web hit dead links, empty results, blocked pages and timeouts regularly. If failures are billed, the true cost per useful page is higher than the sticker price. With success-only billing, the list price is much closer to the effective price.

A quick cost model

Some simple arithmetic shows the scale:

• 100,000 requests: $75
• 1,000,000 requests: $750
• One agent task with 1 search and 3 page fetches (4 requests): $0.003

For most agent products, token costs for the model will still dominate the bill. The retrieval layer is a smaller line item, but it grows linearly with how aggressively an agent browses, so it needs to be tracked.

Where this fits in DeepSeek's strategy

DeepSeek built its reputation on cheap, capable models. Our earlier post, DeepSeek 4.1 Flash vs Claude Opus, looked at that side: similar day-to-day results at a much lower model cost. This launch is different in kind. It prices the tooling around the model rather than the model itself.

This is speculation, but the move would make sense if DeepSeek wants developers to run an entire agent stack, model plus web access, under one account and one invoice. Agent builders otherwise have to combine a model provider with a separate search vendor and often a scraping or cleaning step.

What changes for different builders

Developers building agents. Markdown output should cut down on HTML cleanup before content goes into a prompt or a vector store. It may also use fewer tokens than raw HTML, though that depends on the page and is worth measuring on your own data.

Teams with compliance requirements. Search queries can contain sensitive context. Before sending production traffic, check where requests are processed and how long data is retained. Those details will decide adoption in regulated industries more than the price will.

Hobbyists and small startups. Free failures and a low per-call price make experiments predictable. A prototype making tens of thousands of calls a month would have a retrieval bill in the low double digits of dollars.

What to verify before adopting

Three things are worth testing on your own workload: rate limits under concurrent agent runs, result quality for non-English queries, and how consistently Fetch extracts clean content from JavaScript-heavy pages. The price is clear. Those three factors will determine whether it works as a drop-in replacement for an existing search vendor.

Takeaway

DeepSeek now sells live web access for agents at $0.75 per 1,000 successful requests, billed only when a call returns real results. It is a small line item next to model tokens, but it extends the company's low-cost approach from models to the surrounding agent tooling.

Reference: https://api-docs.deepseek.com/guides/search/