DeepSeek's Search and Fetch APIs Cost $0.75 per 1,000 Successful Requests
DeepSeek prices Search and Fetch at $0.75 per 1,000 successful requests, with failed calls and empty results free, but 5 QPS limits and anti-bot pages may constrain agent use.
DeepSeek's Search and Fetch APIs give agents and RAG applications live web access, priced at $0.75 per 1,000 successful requests. Search returns structured results from DeepSeek's in-house engine, while Fetch retrieves a single page as Markdown.
Both use bearer authentication and the same base URL as the DeepSeek API, and both are read-only JSON APIs with OpenAI-compatible errors. Billing is per successful request, not by tokens or result count: failed requests, empty search results and timeouts are not charged.
Search and Fetch are each limited to 5 QPS per account, with Fetch also subject to a per-domain limit. Server-side tool calls through DeepSeek Harness, Claude Code or Codex are free until November 1, 2026 Beijing time.
The bits that matter
- Search accepts a query of 1–128 Unicode characters and a count of 1–50 results, defaulting to 10. It supports operators including site:, -site:, -, -'...', '...', lang:, after: and before:, and can return page content when need_content is true.
- Each Search result carries a rerank_score from 0 to 3, with 1.5 or above considered high quality, plus a 100–200 character snippet, a 600–800 character summary, an authority tier from 0 to 4, and optional content, site name, icon and RFC 3339 date.
- Fetch takes an HTTP or HTTPS URL up to 8,192 characters with a host and no embedded credentials, and returns the page title and Markdown content. A 404 means no usable content was retrieved; a 403 means the security policy rejected the request.
- Billing is $0.75 per 1,000 successful requests for both APIs. A request is billable only on HTTP 200 with a non-empty results array for Search or page content for Fetch; count does not affect the price, and 402 requires topping up before retrying.
- Exceeding the account-level QPS limit, which applies regardless of which API key is used, or Fetch's per-domain limit returns HTTP 429. Search and Fetch have independent quotas, and these limits are separate from model concurrency limits.
Discussion analysis
Analysis details
20 comments analyzed. Analyzed .
The sample selects active discussion branches and includes available parent comments. It may omit parts of the full thread. The analysis included 20 of 29 usable stored comments because of the input limit. Selected themes do not measure community opinion or how common a view is.
Fetch success against bot walls is uncertain
The Fetch API's ability to return content from pages protected by Cloudflare or Anubis depends on the service and the page. One view holds that US providers like Claude and ChatGPT send identifiable user agents and do not evade bot protections, while others use browser user agents, datacenter or residential IPs. A scraping-service builder reports that a patched Firefox fork plus clean residential proxies covers most cases but still needs 11+ patches and has edge cases. DeepSeek's Search may rely on its own index and cached results, so live-page access is a separate problem.
Default bot evasion raises consent concerns
When a site blocks an attributed fetch tool, Claude and ChatGPT may call curl or write Python with a different user agent to bypass the block, according to a reported pattern. That behavior is described as unethical even if a user explicitly requested it, and more troubling as a default because it overrides site owners' wishes. Anubis is cited as one way to limit endless AI request floods, and a site operator reports residential proxy traffic that dwarfs real traffic while saying the attribution to Chinese AI operators cannot be confirmed.
Scraping arms race burdens ordinary users
The escalation of bot evasion and countermeasures forces ordinary users through Cloudflare or Anubis checks on Apple forums, health insurance portals, county tax sites, electronics distributors and code forges. The bypass work is described as wasting time and money for everyone else. The traffic burden is hard to attribute with certainty, but operators report blocking all countries outside their target area to contain it.
Hosted search trades cost for reliability
Self-hosted SearXNG-based tools and scrapers are free and need no API key, but organizations that need predictable latency and scrape reliability for deep research jobs with many searches and fetches choose hosted providers. A chat platform offers first-party MCP servers for web search and access, yet some customers prioritizing quality over cost use dedicated providers such as LinkUp. The tradeoff is paying for reliability that free scraping does not guarantee.
Data handling questions cut across providers
Organizations choose hosted search providers partly for data processing guarantees, with the do-it-yourself scraping route described as a big leak vector for company secrets. The question is how a hosted provider protects secrets better than calling DeepSeek directly, with local hosting argued as best practice. A counterargument questions whether OpenAI and Anthropic are more trustworthy, while another reply asserts evidence of DeepSeek using stolen credit cards and routing conversations to third parties and asks for comparable evidence against US providers.
DeepSeek's fit depends on guided work
A self-reported daily user credits DeepSeek Flash's search as a differentiator, but claims that it is neck and neck with state-of-the-art models are disputed. It is judged not state of the art for vibe-coding a whole app, though praised for guided work. A question is whether anyone actually one-shots whole apps, and announcements are criticized for focusing on Minecraft-clone demos. Market competition with Anthropic and OpenAI is also debated.