DeepSeek search and fetch – $0.75 per 1000 successful requests
api-docs.deepseek.com
October 18, 2026
7 min read
60/100
Summary
DeepSeek’s Search and Fetch APIs provide agents and retrieval-augmented generation applications with live web access through raw JSON endpoints at api.deepseek.com. Search uses DeepSeek’s in-house search engine to return structured results for queries, while Fetch retrieves one HTTP or HTTPS page and returns its title and content as Markdown. Both endpoints use bearer authentication with a DeepSeek API key. Search accepts queries from 1 to 128 Unicode characters and can return 1 to 50 results, with optional localization, the originating user question, and page content. Queries support site, language, exact-phrase, exclusion, and date operators. Results include relevance scores, snippets, model-oriented summaries, optional content, publication dates, and authority tiers. A rerank score of 1.5 or higher is considered high quality, while authority tiers of 3 or above indicate highly authoritative sources. Search and Fetch cost $0.75 per 1,000 successful requests. Search is billed only for non-empty results, and Fetch only when page content is returned. Server-side calls through DeepSeek Harness, Claude Code, or Codex are free until per-request billing begins at 00:00 Beijing time on November 1, 2026. Each API has a separate account-level limit of 5 requests per second; Fetch also has per-domain limits.
Key Takeaways
- DeepSeek’s Search API returns structured web-search results, and its Fetch API converts a single retrievable web page into Markdown content.
- Search supports up to 50 results per request and accepts operators for sites, languages, phrases, exclusions, and publication-date ranges.
- Both APIs cost $0.75 per 1,000 successful requests, with no charge for failed requests, timeouts, empty searches, or fetches that return no content.
- Search and Fetch each have an account-level limit of 5 requests per second, and Fetch additionally enforces per-domain request limits.
- Server-side Search and Fetch calls made through DeepSeek Harness, Claude Code, or Codex remain free until November 1, 2026 at 00:00 Beijing time.
What the discussion said
The thread treated this less as a consumer search launch than as plumbing for AI agents: can a cheap hosted search-and-fetch layer give models current, usable web context without every developer maintaining a scraping operation? The $0.75-per-thousand price drew real excitement because it undercuts familiar API prices by a wide margin and could make retrieval-augmented agents cheaper to deploy. Several readers also saw DeepSeek Flash as unusually capable for guided work, with search as a meaningful differentiator even if it is not the best tool for autonomous full-app generation. The optimism was tempered by hard operational questions. Self-hosted metasearch sounds free, but commenters argued it degrades quickly under rate limits, CAPTCHAs, anti-bot defenses, and shifting upstream access; enterprises pay for predictable latency, privacy handling, and stable results. Yet cheap pricing alone does not establish quality: premium semantic retrieval and specialized indexes may deliver cleaner, richer context, while DeepSeek’s actual coverage, freshness, fetch behavior, and anti-bot handling remain unclear. Some also objected that AI fetch services can extract and repackage publishers’ work without compensating creators. The central mood is impressed by the price and potential, but waiting for evidence that the service is reliable enough to replace established search infrastructure.
Where opinion split
The sharp dispute is whether DeepSeek’s low price makes it a genuine replacement for premium search and retrieval APIs. Enthusiasts see agent-grade web access becoming a commodity, while skeptics argue that reliable coverage, anti-bot resilience, semantic signal quality, and enterprise guarantees are exactly what the higher-priced providers sell.
Community Sentiment
Positives
- At $0.75 per thousand successful requests, DeepSeek makes live web context cheap enough that retrieval-heavy agents no longer need a premium-budget excuse.
- Hosted search saves AI builders from the brittle treadmill of proxies, browser automation, CAPTCHAs, and upstream rate limits that turns supposedly free scraping into unreliable agent input.
- Several readers rate DeepSeek Flash highly for guided daily work, and see integrated search as the feature that helps it punch above its price.
- Aggressive pricing could force richly funded AI-search vendors to prove their semantic retrieval and indexing advantages rather than rely on fat API margins.
Concerns
- Cheap search is not automatically good agent context; commenters question DeepSeek’s result quality, freshness, coverage, and ability to fetch pages behind modern bot defenses.
- Self-hosted and scraped metasearch often feeds models inconsistent or degraded results once providers throttle requests or deploy anti-automation challenges.
- DeepSeek Flash is considered useful for guided tasks but clearly behind top proprietary models when asked to autonomously build an entire application.
- Fetching and summarizing web pages for AI agents may further weaken incentives for publishers to create material when the service does not share revenue with them.
