DeepSeek Search Is No Longer Free After Tomorrow

On October 17, 2026, DeepSeek announced two new API endpoints: Search, which returns ranked web results, and Fetch, which retrieves a single page as Markdown. Both launched open to all API users and — for server-side calls originating from DeepSeek Harness, Claude Code, and Codex — at zero cost. That free window closes at 16:00 UTC on October 31 (00:00 Beijing time on November 1). DeepSeek priced both endpoints at $0.75 per 1,000 successful calls.

The announcement matters for any team that has wired these endpoints into an AI agent’s tool-use stack, whether directly or via a framework. Failed requests, empty results, and timeouts are not billed — only calls that return HTTP 200 with results. The number of results requested (up to 50 per Search call) does not change the price.

What the Two Endpoints Do

The Search endpoint takes a query and returns up to 50 web results. Each result includes a title, URL, publication date (when known), a snippet for display, a longer model-readable summary, an authority tier (0–4), and a relevance score (0–3). It supports standard search operators: site:, -site:, lang:, after:, before:, and quoted phrases. Authentication is the same bearer token used for DeepSeek’s model API — no second key.

The Fetch endpoint takes a URL and returns the page as Markdown, stripping navigation, ads, and boilerplate. Rate limits are 100 requests per second for Search and 25 QPS for Fetch (plus an undisclosed per-domain cap). Hitting the cap returns HTTP 429; an insufficient balance returns HTTP 402. DeepSeek’s own claim is that its engine serves hundreds of millions of requests daily — that claim is from DeepSeek itself, not from an independent audit.

Cost Math at Agent Scale

The per-call price is low enough that most small pipelines will absorb it without budget review. Larger deployments are a different story.

The $135/month row is the cost example DeepSeek’s own documentation uses. At $1,350/month the provider selection decision becomes real. That is when it is worth benchmarking result quality against alternatives — price per call is only part of the equation.

The Search API Price Map

Here is the current market as of October 29, 2026, based on published list prices. All figures are per 1,000 successful calls and exclude any model token charges layered on top.

Sources: ppc.land (DeepSeek); Parallel.ai (Cloudflare); tokely.me (Brave, Tavily, OpenAI, Perplexity). Prices as of October 29, 2026 — check provider pages before budget decisions.

For Swiss & EU teams

DeepSeek’s privacy policy states that all data is stored on servers in the People’s Republic of China, governed by Chinese law including the National Intelligence Law. The company offers no enterprise DPA, which means there is no Article 28 controller-processor agreement and no mechanism to transfer EU personal data under GDPR Chapter V. Italy’s privacy regulator has already blocked DeepSeek’s consumer products; Luxembourg’s CNPD issued specific guidance. Switzerland’s nFADP (FADP) applies the same adequacy logic: China has no adequacy decision from Switzerland’s Federal Council.

The practical implication for the Search API: any query containing personal data — a customer name, an email address, a project code that maps to a person — is a cross-border personal data transfer to China. For teams under FADP or GDPR, this is a non-starter without a legal basis that currently does not exist. Brave Search (independent EU-friendly index, documented GDPR posture) or Cloudflare’s Web Search API with the Ceramic.ai or Linkup providers (Zero Data Retention available) are the lower-friction alternatives at this price tier.

What to Do on Monday

1. Audit your agent pipelines today. Search your codebase and infra configs for calls to api.deepseek.com/v1/tools/search or /v1/tools/fetch. If you are calling these from Claude Code, Codex, or DeepSeek Harness, billing starts at 16:00 UTC on October 31 — those are the calls that were free.
2. Check your DeepSeek API balance. A zero balance at billing start returns HTTP 402 and silently breaks agent calls. Log in to platform.deepseek.com and either top up or wire in a usage alert before October 31.
3. For EU/FADP-regulated workloads, reroute now. Switch Search calls to Brave Search API or to Cloudflare’s Web Search API with the Ceramic.ai or Linkup provider (both offer Zero Data Retention). Ceramic.ai at $0.25/1K is the cheapest confirmed option on the market today — it is lexical-only, so test it on your query patterns before committing.
4. Run a 48-hour cost projection. Take your current search call volume, multiply by $0.00075, and compare to Brave at $0.005. At fewer than 10,000 calls a month the difference is negligible; above 100,000 calls a month you have a meaningful decision to make.
5. Read the model context too. The Search and Fetch APIs launched alongside DeepSeek V4.1-Flash-1017, which also hit 77% on DeepSWE — the model and the search tooling are part of the same October 17 release package.

Further Reading

• DeepSeek prices web search for AI agents at $0.75 per 1,000 calls — detailed breakdown of the endpoint specs and what the billing change actually covers.
• Cloudflare Web Search API: providers, pricing, and how it compares — the best current overview of Ceramic, Exa, and Linkup features and prices via Cloudflare’s unified API.
• How to Secure Your AI Coding Agent Pipeline (2026) — if you are auditing your DeepSeek search calls you should also audit what those agents can access: MCP governance, dependency injection, and NIS2 notes included.

Your turn: Which search API does your team’s agent pipeline currently use, and did the DeepSeek free period change your default choice? Reply to our newsletter or send us a note — we feature the best answers in the Friday Scorecard.

How this article was made: AI researched and wrote this article under the Christian persona, using the sources linked above, and it was published automatically without a human edit. Editorial guidelines are set by Adi. Spotted an error? Tell us and we will correct it.