DeepSeek launched Search API and Fetch API for AI agents, pricing web search and page retrieval per successful request

Decision brief

What changed

DeepSeek opened two new JSON endpoints for external developers: a Search API for live web results and a Fetch API for page content retrieval, both using the same bearer authentication as its main API. DeepSeek set pricing at $0.75 per 1,000 successful requests for each service, with separate QPS limits and billing scheduled to begin on November 1, 2026.

Why it matters

This gives companies building AI agents or RAG workflows a new low-cost retrieval option that can reduce the infrastructure cost of adding live web search and page reading to applications. For decision-makers, the relevant shift is not just lower list pricing, but the billing model: charges apply per successful request rather than by tokens or result count, which can simplify unit economics for agent workflows while still leaving downstream model-token costs in place. The launch also broadens DeepSeek from model access into a more complete agent tooling stack, which may affect vendor selection for teams optimizing for cost and integration simplicity.

Affected roles

CEO CFO CTO COO

Evidence

All four cited outlets report the launch of DeepSeek’s Search and Fetch APIs for live web retrieval. The Neuron provides the most specific operational details on price, billing start date, per-successful-request charging, and QPS limits, while The New Stack independently confirms external availability and MarkTechPost supports the cost-positioning theme by noting retrieval can be cheap while token-processing costs still remain.

What remains uncertain

The coverage does not verify real-world reliability, relevance quality, geographic coverage, latency, or how the service performs at scale versus alternatives; the ~85% cheaper framing is reported by The Neuron rather than independently benchmarked across identical workloads in the provided coverage. It is also unclear how much total agent cost will actually fall for a given business, because retrieved-page token volume, repeated agent loops, and any free-period usage limits could outweigh retrieval-call savings.

Monitor next

Watch for post-launch developer signals after billing begins on November 1, 2026—especially disclosed usage levels, reliability metrics, and independent cost/performance comparisons against other search and retrieval APIs.

Analytical support, not advice — assumptions and open questions stated above.

Source coverage