DeepSeek peak/off-peak price clock

The DeepSeek API costs half as much during off-peak hours. The windows are fixed in UTC, which makes them awkward to keep track of anywhere else in the world – so this page tells you where you are right now, when it flips next in your own time zone, and what both models cost either way. Queue the non-urgent jobs for the cheap half and the bill drops by 50%.

Peak windows only apply on weekdays (Monday–Friday, Beijing time). Weekends are off-peak all day, so Saturday and Sunday never see a surcharge.

Per 1M tokens. Input is cache miss. Cache hits cost $0.003 off-peak ($0.006 peak) for Flash and $0.022 off-peak ($0.044 peak) for Pro.

DeepSeek dropped the planned Sept 14 retirement, so V4 Pro stays on its own rates. The legacy names V4 Flash and Vision Exp are still accepted and served by V4.1 Flash at Flash rates. Images are converted into tokens by dimensions and billed as input together with text. Both models take 1M context with up to 384K output. Concurrency is 2500 for Flash and 500 for Pro.

Rates are from the official DeepSeek pricing page. Image billing follows the Vision token rule and limits follow the concurrency table.

Rather not watch the clock? DeepSeek models run on OpenCode Go at a flat rate, and signing up through my OpenCode referral link gets us each a $5 reward.

Some links here are affiliate or referral links. Buying through them supports this site at no extra cost to you, and it doesn't affect my opinions.