DeepSeek search and fetch – $0.75 per 1000 successful requests

Original Article ↗ Hacker News Discussion ↗

DeepSeek has launched a low-cost web search and fetch API aimed at AI agents, undercutting many existing providers and prompting comparisons with services like Exa, Brave, Kagi, and SerpAPI on both price and quality. Commenters debate the technical challenges of bypassing bot protections and CAPTCHAs, as well as privacy, data retention, and the heavy crawler traffic that AI-driven scraping imposes on small sites. The thread also reflects broader concerns about how cheap, automated access to web content shifts power in search, strains infrastructure, and may erode incentives for content creation.

Scope of the DeepSeek Search/Fetch Offering

  • API is aimed at agents, not humans, to perform web search and page fetches.
  • Priced at $0.75 per 1,000 successful requests; compared to other APIs, it’s on the cheap end (others mentioned: Exa, Brave, Kagi, SERP API, Ceramic, LinkUp, Uruky, etc.).
  • Some users report the actual search/fetch quality as mediocre compared to specialized providers and suggest using other tools for higher-quality or semantic retrieval.

Bot Evasion, Anti‑Bot Arms Race, and Impact on the Web

  • Many comments focus on how these services bypass Cloudflare, Anubis, CAPTCHAs, and bot detection.
  • Techniques discussed: remote‑controlled hardened Firefox/Chrome builds, residential proxy networks, curl impersonation, UA spoofing.
  • Site operators complain that AI-driven scraping massively outnumbers real traffic, ignores robots.txt and cache headers, and forces them behind CAPTCHAs, VPNs, or services like Anubis.
  • Strong criticism that such tooling escalates an arms race and degrades the web for regular users; defenders argue it is self‑protection against exploitative, ad‑driven web practices.

Privacy, Logging, and Data Retention

  • Concern that DeepSeek and similar services log target URLs and usage, effectively storing user browsing/search histories.
  • Some argue for minimal logging (only billing metadata) and criticize widespread data hoarding; note that a few providers advertise optional or no-retention modes, often at higher cost.
  • Zero‑data‑retention is a hard requirement for some use cases (e.g., verifying scientific citations), making DeepSeek unsuitable for them.

LLM Quality, Use Cases, and Economics

  • DeepSeek Flash is seen as highly cost‑effective and close to state‑of‑the‑art for many coding and agentic tasks, though not generally superior to frontier models.
  • Debate over whether open or Chinese models can undercut US incumbents by combining strong models with ultra‑cheap search.
  • Some see this pricing as unsustainable and expect future increases; others say DeepSeek can afford low margins due to backing from a hedge-fund parent and/or broader strategic motives.

Alternatives, Self‑Hosting, and Future of Search

  • Several users roll their own MCP/tools using third‑party scrapers or metasearch (e.g., SearxNG), but report heavy rate limiting, CAPTCHA walls, and inconsistency.
  • Hosted search is seen as necessary for reliability, latency, and large‑scale agent use; self‑hosting full web indexes is viewed as costly and complex.
  • Some speculate search APIs could eventually become subsidized or even “negative cost” to capture agent-driven traffic and SEO leverage.