Off-peak pricing: half price for timing
The single cheapest optimisation available. No code change, no quality loss — just scheduling.
The window
| Period | UTC | Beijing | Rate |
|---|---|---|---|
| Peak | 01:00–04:00, 06:00–10:00 | 09:00–12:00, 14:00–18:00 | 2.0x |
| Off-peak | All other hours | All other hours | 1.0x |
Peak applies Monday to Friday only. Weekends and Chinese public holidays are entirely off-peak.
Why it is shaped that way
Converted to Beijing time, peak hours are 09:00–12:00 and 14:00–18:00 — the Chinese working day. That is deliberate: the discount exists to move load out of business hours, not to reward geography. If you are in Europe or the Americas, most of your working day may already be off-peak.
What to move
| Workload | Movable? | Action |
|---|---|---|
| Batch data cleaning, offline analysis | Yes | Schedule to night or weekend |
| Bulk content generation | Yes | Queue and drain off-peak |
| Embedding backfills, re-indexing | Yes | Nightly job |
| Interactive chat | No | Accept the peak rate |
| Real-time user requests | No | Use caching instead |
Two more levers
1. Cache hits
A cache hit costs 1/50th of a cache miss. Put stable content — system prompts, knowledge-base prefixes, long documents — at the front of the request so it gets cached across calls. This works regardless of the time of day.
2. Provider choice
The same open-weight model can differ by up to 18x across providers. See provider comparison. Off-peak from the official API is not necessarily cheaper than a third-party provider at peak.
Estimate the saving
Take your current monthly bill. If all of it currently lands in peak hours, moving the batch portion (typically 40–70% of volume) saves 20–35% with no other change.
Run the numbers in the cost calculator — set the peak-hour share to 0 and compare.
Related
Full price list · Cost calculator · Price history · Provider comparison