Alibaba’s current subscription product is a credits-based Token Plan. The older version of this article described a roughly 90,000-request coding allowance, which is no longer a safe description of the live plan. Current team plans assign seats and draw model usage from a credits pool.
How billing differs
Token Plan and metered API
| Token Plan | Subscription; credits pool; seats; dedicated plan key and endpoint |
|---|---|
| Pay-as-you-go | Input, output, and cache usage billed at the selected model's regional rate |
| Best fit for plan | Regular team usage that fits the included credit allowance |
| Best fit for metered | Light, irregular, or tightly measured workloads |
The plan supports multiple Qwen and third-party models, but each model consumes credits differently. A cheaper model can stretch the pool further than the flagship. Check the purchase page for current seat prices and credit multipliers because promotions and supported models change.
Endpoint warning
Estimate with real usage
Run a typical week on pay-as-you-go billing. Record input, output, and cache charges from the cost dashboard, then compare the monthly estimate with a plan seat and its expected credit use. Include automated agent loops because retries and tool calls can use much more than a short chat.
Pay-as-you-go also makes model-level costs easier to see. For example, Qwen3.8-Max has regional list prices and separate cache rates. A Token Plan combines usage into credits, which is simpler for budgeting but less direct when you compare individual calls.
Keep the key types separate
Name environment variables clearly and avoid reusing one key across both routes. When a coding tool returns 401 after a plan purchase, verify the endpoint first, then the key and workspace region.
For current flagship costs, see Qwen3.8-Max pricing. For setup, see Qwen3.8-Max with Claude Code.
