DeepSeek launches search API, unveils V4.1 flash snapshot, and releases harness v0.2.1-rc1

On October 17, 2026, DeepSeek announced the official launch of its Search and Fetch APIs, both powered by its proprietary search engine. These APIs provide real-time web content for Agent and RAG applications, with Search returning structured results and Fetch delivering Markdown-formatted page content. Both are JSON-based, with a base URL of api.deepseek.com and Bearer token authentication. Pricing is $5 per 1,000 successful requests, with no charges for failed or empty results. The Search API is free for the first two weeks after its launch, excluding searches initiated via Claude Code or Codex. A user account on X, AI Tracker Bot, detected a temporary API entry labeled 'deepseek-v4.1-flash-1017,' which was later removed. This suggests a potential V4.1 Flash snapshot, though DeepSeek has not confirmed its existence. Meanwhile, the company also released v0.2.1-rc1 of DeepSeek Harness, featuring new plugins, improved tooling, and compatibility layers for external models. The update includes experimental features like machine translation and Git Worktrees support, alongside enhanced user interface controls for model configuration and session management.