DeepSeek just made a move that quietly saves everyone money: starting Sunday, August 23, 2026, off-peak rates apply all day on weekends.
That means Saturdays and Sundays are now half-price — all day, every day. Here's the full picture.
What changed
Effective 00:00 Beijing Time, August 23, 2026, DeepSeek adjusted its peak/off-peak billing rules. The new rule is simple:
Weekends (Saturday and Sunday, Beijing Time) = off-peak rates all day.
Previously, peak hours were only defined per day (01:00–04:00 and 06:00–10:00 UTC), so weekend work could still land in a peak window. Now the entire weekend is discounted.
What the rates look like
DeepSeek's off-peak prices are exactly half of peak. For deepseek-v4-flash-vision-exp (and flash):
| Peak | Off-peak | |
|---|---|---|
| Input (cache miss) | $0.44 / M | $0.22 / M |
| Input (cache hit) | $0.014 / M | $0.007 / M |
| Output | $1.32 / M | $0.66 / M |
So on a weekend, you're paying half for every token. That's a real cut for anyone running agent workloads.
Why this matters
Agents burn tokens around the clock — scheduled jobs, background tasks, long conversation loops. The distinction between peak and off-peak used to matter mostly for weekday batch runs. Now the whole weekend is the cheap window.
For anyone running an AI agent, this changes the math:
- Batch jobs, reports, and heavy generation — run them on the weekend, pay half.
- Weekday evenings (after 10:00 UTC) were already off-peak; weekends are now the same everywhere.
The practical takeaway
This is a deliberate nudge toward weekend computing — schedule your token-heavy work for Saturday/Sunday and DeepSeek cuts your bill in half. For a 1M-token weekend batch that's the difference between $0.44 and $0.22 per million input tokens.
Worth recalibrating your cron jobs around it.
Source: DeepSeek Models & Pricing, August 2026.
