The short answer
Alibaba's Token Plan Individual measures supported model and Harness-tool use in Credits. Every successful call consumes a variable number of Credits and counts against two limits at once: a five-hour limit and a seven-day limit. The service pauses when either limit is reached, then resumes when that fixed window ends and its quota resets.
The official documentation calls these fixed windows, not a continuously rolling balance. That distinction matters. The timer begins with the first invocation in a new window, and the full allowance resets at the end. Credits do not return gradually as individual calls become five hours old. The Coding Plan uses a different replenishment mechanism, so language used for one plan should not be assumed to describe the other.
What a Credit represents
A Credit is the plan's common usage unit; it is not the same thing as one prompt, one API request, one token or one US dollar. Alibaba says consumption is dynamically determined by the model, token usage, thinking mode and tool calls. A short request to an efficient text model can therefore consume differently from a long reasoning session, a web-search-assisted answer or a media-generation task.
Both quota counters receive the same deduction. If a call consumes 100 Credits, the remaining allowance in the active five-hour window and the active seven-day window each falls by 100. The actual deduction is shown in the subscription usage details, which is the appropriate place to evaluate a real workload. A generic conversion such as 'one Credit equals one message' would be misleading.
How the two fixed windows work
This design places a short-term ceiling on bursts and a second ceiling on sustained weekly usage. A developer can have a fresh five-hour allowance but remain paused because the seven-day total has been exhausted. Conversely, the weekly counter can have room while a concentrated session reaches the five-hour limit.
- Five-hour window: the first invocation starts a five-hour timer. Consumption accumulates until the timer ends. If the allowance is exhausted sooner, access pauses until that end time.
- Seven-day window: the first invocation starts a seven-day timer. It is not tied to Monday or another calendar reset day.
- Independent enforcement: remaining seven-day Credits do not override an exhausted five-hour allowance, and remaining five-hour Credits do not override an exhausted seven-day allowance.
- No carry-over: unused quota in either completed window does not move into the next one.
A Standard-plan example
Standard currently has a 3,000-Credit five-hour limit and a 10,000-Credit seven-day limit. Suppose the first call is made at 10:00 on 20 July. That starts a five-hour window ending at 15:00 and a seven-day window ending on 27 July.
If 2,000 Credits are used at 10:00 and another 1,000 at 11:00, the five-hour allowance is exhausted. Service remains paused until 15:00, even though the seven-day window may still contain 7,000 Credits. At 15:00, the five-hour allowance resets in full. Consumption after that continues to count against the same seven-day window until its own end.
This example explains timing, not how many coding tasks a plan guarantees. Credit deductions vary by request, and agent tools can make several model or tool calls while completing one user-visible task.
Current limits by tier
Alibaba also publishes recommended concurrent-agent ranges of one to two for Lite, three to four for Standard and six to eight for Pro. These are plan guidance, not a promise that every combination of model, task and network conditions will deliver identical throughput. Peak periods may involve queuing.
- Lite: 700 Credits per five-hour window and 2,500 Credits per seven-day window.
- Standard: 3,000 Credits per five-hour window and 10,000 Credits per seven-day window.
- Pro: 12,000 Credits per five-hour window and 40,000 Credits per seven-day window.
What happens at the limit
When either quota is exhausted, Token Plan Individual blocks further calls rather than moving them automatically to pay-as-you-go billing. The documented options are to wait for the relevant reset or upgrade to a higher tier. An upgrade is prorated for the remaining subscription period and raises both limits immediately.
Renewing early does not add quota to the current fixed windows; it extends the subscription period. Individual currently does not support cancellation, although auto-renewal can be disabled. Those terms make it sensible to confirm the current plan page before purchase and to start with measured interactive use rather than assuming a monthly message count.