Update: DeepSeek's peak/off-peak API pricing starts Aug 16
Peak covers 7 of 24 UTC hours and pushes V4-Pro output to $3.96/M; off-peak is half. Plus faster Cursor cloud agents and GitHub's Agent Plugins 1.0.
Copy markdown
Seven peak hours, all in UTC
From Aug 16, peak billing runs 01:00–04:00 and 06:00–10:00 UTC — 9am–noon and 2–6pm Beijing. Every other hour is off-peak. If your agents run on a schedule, that seven-hour band is the one to route around.
Off-peak is exactly half price
Off-peak rates are half the peak rates. V4-Pro output runs $1.98/M off-peak vs $3.96/M at peak; cache-miss input is $0.66 vs $1.32. Shift batch and overnight jobs into off-peak and you pay 50% less for identical tokens.
It's a hike from today either way
Until the switch, V4-Pro is a flat $0.435/M input and $0.87/M output. Even the new off-peak floor sits above that, and peak lands at roughly 3–4.6x today's rate — cache-hit reads jump too. Re-check your cost model before the 16th.
Elsewhere: Cursor cloud agents boot 3x faster
Cursor's new Builds pre-bakes your dev environment in the background, so cloud agents launch in a ready-to-use copy — 3x faster time-to-first-token. A Builds tab tracks status, logs, commit SHAs, and deploy history.
Elsewhere: GitHub ships Agent Plugins 1.0
GitHub's Agent Plugins 1.0 now run across VS Code, the Copilot CLI, and the Copilot app — build one agent extension and it works in all three surfaces instead of one.