DeepSeek opens its app toolkit via API with deepseek-flash at fractions of a cent

DeepSeek now sells the building blocks behind its own chat app. Listed tools include web search, image search, deep-thinking iterative search, file and PDF parsing, URL fetching, safety checks, and standalone OCR, ASR and TTS endpoints. Voice tools launch at a 50% discount. DeepSeek also open-sourced a chat client that can import history from its consumer app.
Everything runs through deepseek-flash, a V4.1 Flash checkpoint. The endpoints follow OpenAI- and Anthropic-style protocols, so swapping them into existing agent frameworks is mostly a configuration change.
Pricing is the story. Developers compared $0.75 per 1K searches against roughly $10 per 1K at OpenAI. One widely shared X post argued that "tool calls are becoming fractions of a cent utilities." That squeezes search-API startups and the tool layers of Western labs alike. It also lands as AWS adds its own serverless web search to AgentCore.
The caveats: data-residency and compliance concerns limit enterprise adoption of China-hosted APIs, and DeepSeek faces separate reports that V4 trained on export-barred Blackwell chips. Search quality and latency outside China are also unproven.