The Overnight

The Overnight — 2026-10-18

DeepSeek refreshes V4.1-Flash and opens a search API at 5 yuan per 1,000 calls

What China's AI world did while the West was asleep. · 2026-10-18

Share: Share on X LinkedIn

The Number

  • DeepSeek prices its new Search and Fetch APIs at 5 yuan per 1,000 calls each, its first paid web-retrieval tier, per DeepSeek docs as of 2026-10-17.
  • Qwen3.8-max runs 12 yuan per 1M input tokens and 36 yuan output on DashScope, with the Prime variant at 24/72, per Alibaba's pricing page as of 2026-10-17.

Top of the Fold

  • DeepSeek updates deepseek-flash to the V4.1-Flash-1017 build across web, app and API, with weights on HuggingFace and ModelScope, a month after the first Flash release, per 智东西.
  • DeepSeek folds its retired v4-flash names into the new 1017 model and bills legacy calls at the Flash price, keeping the API surface unchanged for developers, per DeepSeek docs.

Untranslated

  • DeepSeek ships a self-built search engine behind new Search and Fetch APIs, moving from a model vendor toward owning the retrieval layer, per DeepSeek docs.
  • DeepSeek also puts the DeepSeek Harness into worldwide developer preview for agent-harness builders, a quieter release than the model refresh, per DeepSeek docs.
  • Otherwise quiet across the Chinese community overnight.

The Kicker

  • Apple's first foldable iPhone Duo sold out in roughly 1 minute at its 8pm China preorder, pushing official shipping to late January 2027 and a 2TB unit to a $5,000 eBay flip, per 极客公园.