Schedule around resets
Monthly, five-hour and weekly resets, off-peak windows for batch work, and concurrency limits.
New plans: the monthly reset
Included credits reset monthly, on the day of the month your subscription started. Annual plans follow the same rule. A subscription that started on the 3rd resets on the 3rd.
Included usage does not roll over. Unused credits end at the reset, so pace the month rather than spending it all in its last week. Free accounts reset monthly from the day they signed up.
Paid plans get an email at 90% of the monthly usage. Treat it as an early warning, and read When capped for what happens after that.
Old plans: five-hour blocks and a weekly window
Old-plan sessions are fixed five-hour blocks that end at minute :55. The blocks do not drift. But five and twenty-four share no common factor, so the boundaries walk through every clock hour over five days.
The weekly window is fixed at seven days. We saw one reset land exactly seven days after the previous one. In our checks, the session boundaries showed no drift across six observed resets in August 2026.
For a heavy run, start just after a block boundary. Then the run has almost a full block ahead of it, not the tail of one.
Off-peak windows for batch jobs
Off-peak pricing applies outside 12:00 to 18:00 UTC on weekdays, and all day at weekends.
- Weekdays, off-peak: 00:00 to 12:00 UTC and 18:00 to 24:00 UTC.
- Weekends, off-peak: all day, UTC.
Only models with an off-peak row get the lower price. At the last check that was deepseek-v4.1-flash and deepseek-v4-pro, at half their standard price. For a model without an off-peak row, the window changes nothing, so check the atlas before you move a job.
A batch job that can wait belongs in those windows. A job that runs at 13:00 UTC on a Tuesday pays standard prices on deepseek-v4.1-flash. The same job at 19:00 UTC pays the off-peak price.
Concurrency and queueing
| Plan | Concurrent requests |
|---|---|
| Free | 1 |
| Pro | 3 |
| Max and Team | 10 |
Requests over the limit are queued. If the queue is full, the request is rejected until a slot opens. Set your worker count to the plan’s limit. Extra workers mostly add queue time and rejections.
Check the real numbers
Paste an Ollama Cloud key on the home page. Exact monthly dollars on new plans, estimates on old ones, nothing stored.
Check your own usage