Skip to main content

Usage and quota

Token Plan Credits billing rules, deduction order, reset rules, quota exhaustion, and usage troubleshooting.

Credentials and billing boundary

  • Token Plan dedicated API key (sk-sp-): Used for Token Plan plan calls; deducts plan quota and usage packs.
  • Alibaba Cloud Model Studio general API key (sk-): Used only to deduct the Harness benefits quota included with Personal Edition; it cannot be used to deduct model call quota.
  • Free quotas, resource packs, savings plans, and account balances: Belong to the Alibaba Cloud Model Studio pay-as-you-go system and are independent of Token Plan quotas.

Deduction order

  1. The plan quota for the current subscription month is deducted first. For Team Edition, the monthly quota of the current seat is deducted first.
  2. After the plan quota is used up, Personal Edition automatically deducts from usage packs; Team Edition deducts from shared usage packs. When multiple shared usage packs are held, the pack that expires soonest is deducted first.
  3. After all available quotas are used up, calls that consume Credits are blocked and are not automatically switched to pay-as-you-go. You can wait for the next subscription month to reset, upgrade the plan or seat, or purchase usage packs or shared usage packs.
Harness benefits are not part of the deduction order. For details, see Harness benefits.

Quota rules

Reset and carryover

  • The quota resets automatically per subscription month. At the start of each subscription month, the quota is restored to the full monthly quota of the corresponding tier. The reset time depends on the subscription date, not the 1st of each month.
  • Unused quota within a subscription month does not carry over to the next cycle.

Upgrade quota calculation

Upgrading a plan or seat does not reset the current subscription cycle: you pay the price difference for the remaining days of the current subscription cycle, additional Credits for the current cycle are granted only based on the quota difference, and the new tier's monthly quota applies from the next subscription month. The Credits granted by an upgrade (including seat upgrades) are not the full monthly quota of the new tier. They are calculated based on the actual price difference paid and the actual remaining valid time in the current subscription cycle. The formulas are as follows. 
Upgrade price difference = (New tier price − Old tier price) × Remaining days ÷ 30
(Less than one day counts as one day)

Additional quota for the current cycle after upgrade = Remaining days ÷ 30 × (Monthly quota after upgrade − Monthly quota before upgrade)
(Less than one day counts as one day; the result is rounded up)
The actual price difference is subject to the price breakdown shown on the payment page. Team Edition upgrades are performed per seat; the actual price difference and available quota are subject to the console. For example: On September 2, subscribe to the Personal Edition Lite plan (39 CNY/month, 20,000 Credits per month) and upgrade to the Essential plan (79 CNY/month, 45,000 Credits per month) on September 12, with 20 days remaining in the current subscription cycle. The price difference is (79 − 39) × 20 ÷ 30 ≈ 26.67 CNY, and the additional quota for the current cycle is 20 ÷ 30 × (45,000 − 20,000) ≈ 16,667 Credits (rounded up). For how to calculate the remaining Harness benefits quota after an upgrade, see Harness benefits · Quota after upgrade.

Renewal and add-ons

  • Renewal only extends the subscription validity period and does not add quota to the current subscription cycle.
  • The cost and Credits for adding seats are prorated by the remaining duration and expire together with the existing subscription.

Credits consumption rules

Credits are the unified billing unit for Token Plan and are not equivalent to tokens. Different models have different unit prices. The Credits for a single request are determined by input tokens, cached tokens, and output tokens. Using qwen3.6-plus in Team Edition as an example:
Token typeQuantityCredits consumed
Input tokens8,3491.67
Cached tokens40,7940.82
Output tokens5730.69
Total—Approx. 3.18 Credits
Main factors affecting per-request consumption:
  • Model unit price: Higher-priced models consume more Credits for the same number of tokens.
  • Context length: Longer message history means more input tokens.
  • Thinking mode: Enabling thinking increases output tokens.
  • Tool calls: Built-in tools, MCP tools, and subtasks triggered through the model API increase the number of model calls and consume Credits; Harness benefits tools called directly through sk- do not consume Credits.
  • Cache hits: The unit price of cached tokens is usually lower than that of regular input tokens.
Actual consumption per request is subject to Usage Analysis in the console. To estimate consumption, run a short task first, and then check the consumption details in Usage Analysis.

Reduce Credits consumption

Credits consumption is mainly affected by context length, model unit price, thinking mode, tool calls, and task type.

Conversations and context

  • Compress message history or start a new conversation to avoid long context accumulation.
  • Disable thinking mode for simple tasks to reduce output tokens.

Models and tools

  • Use a lightweight model for non-complex tasks, for example, qwen3.8-flash instead of qwen3.8-max.
  • Reduce unnecessary tool calls, MCP calls, and parallel multi-agent runs.
  • Use caching for repeated content. Cached tokens have a lower unit price than regular input tokens.

Video and other multimodal tasks

Multimodal models such as video generation consume more Credits per request than text conversations, and consumption increases with duration and resolution. Asynchronous tasks are settled together after completion, so submitting many tasks at once may quickly reach the quota. For first use, run a shorter, lower-resolution task first.

Night discount

Personal Edition and Team Edition offer Credits discounts for some models from 22:00 to 08:00 the next day. For specific models, discounts, and promotion validity, see Models supported by Personal Edition and Models supported by Team Edition.

Usage monitoring

Check Usage Analysis in the console regularly to identify high-consumption models and members.

Quota exhaustion and recovery

After the plan quota and usage packs (including Team Edition shared usage packs) are both used up, model calls that consume Credits are blocked and are not automatically billed pay-as-you-go. Recovery options:

Usage troubleshooting

For the differences between quota exhaustion, rate limiting, concurrency limits, and authentication errors, see Integration errors and rate limiting. If consumption is faster than expected, check the context length, thinking mode, model unit price, tool calls, and asynchronous task settlement. See How many Credits does a call consume and Reduce Credits consumption.

Third-party tool statistics differ

Token usage shown by third-party tools (such as Claude Code) may be lower than that in the Model Studio console. Common causes:
  • Different measurement scope: Third-party tools show only model input and output tokens, while the console counts the complete set of consumption items.
  • Hidden consumption: System prompts, tool definitions (schemas), project conventions, multi-turn conversation history, tool call parameters and results, and model reasoning content all count toward Credits but are not shown in the tools.
  • Different model unit prices: For the same number of tokens, higher-priced models consume more Credits.

Pay-as-you-go charges

Common causes:
  • Calls before activation: Calls made before Token Plan is activated are billed independently and cannot be deducted from the plan.
  • Configuration errors: Not using the Token Plan dedicated API Key (starting with sk-sp-) and Base URL, for example, using the general Model Studio endpoint dashscope.aliyuncs.com or a Coding Plan key by mistake.
Pay-as-you-go charges already incurred cannot be offset by Credits afterward or refunded. Check the API Key and Base URL immediately to ensure that subsequent calls are deducted correctly.

Usage is 0

  • No actual consumption has been generated.
  • A non-Token Plan dedicated API Key is used.
If usage is 0 but calls are still being made, check whether an Alibaba Cloud Model Studio general API key was used by mistake in the tool and generated pay-as-you-go charges.