DeepSeek V4 Pro & Flash — Launch (Manual Selection)

April 24, 2026

Superseded — April 27, 2026

This is the original launch state. Three days later, DeepSeek V4 Pro and V4 Flash moved to zero-retention endpoints — they're now picked by Auto routing and Swarm, with no privacy warning. Read that note for the current state.

DeepSeek's two newest open-weight models — V4 Pro and V4 Flash, both with a 1M-token context window — became selectable in Alfrada at launch. Because both were initially hosted in China, Auto routing couldn't pick them and a privacy notice appeared above the chat input whenever either was active. Pricing was workhorse-tier for Pro, budget-tier for Flash.

Both models supported reasoning mode and full tool calling.

What changed three days later

The April 27 update moved both models to zero-retention endpoints. The privacy notice, the manual-selection-only state, and the China-only routing are all gone. See the April 27 ZDR update for current behaviour.