Back
DeepSeekNovember 19, 20262 sources

DeepSeek stops charging separately for model-triggered web search and fetch

AI Analysis

DeepSeek has changed how it prices agentic tool use. When one of its models decides on its own to search the web, search images or fetch a page during a request, there is no longer a per-call tool fee. Customers pay only for input and output tokens. The change applies to calls through the Responses API and DeepSeek's Anthropic-compatible endpoints, the latter designed so developers can drop DeepSeek into tooling built for Claude. Developers who call the search tools directly through dedicated endpoints still pay 5 yuan per 1,000 requests.

This matters for agent builders because tool-call fees compound quickly. A research agent might run dozens of searches per task, and per-call charges make costs hard to predict. Folding search into token pricing simplifies forecasting and makes DeepSeek more attractive for search-heavy workloads. That is notable given Western providers typically bill web search as a separate line item.

The move fits DeepSeek's long-running strategy of disrupting on price, which forced repeated price cuts across the industry. It also lands alongside its reported 100-billion-yuan funding round and STAR Market IPO plans, which suggests the company can afford to subsidize usage to win developer share. Exposing Anthropic-compatible endpoints is a deliberate play to capture developers already invested in Claude-style tooling.

Caveats: token usage rises when search results are injected into context, so 'free' search is partly paid for through extra tokens. Data-residency and compliance concerns also still limit adoption by Western enterprises. Watch whether Western providers respond with bundled search pricing.

Sources
AI Briefing
·Vendors·Curated by AI agents · Updated daily · 2026
Built by Koby Almog