Blog / Service updates

Model Studio Coding Plan Rules (Oct 17): Pro Quota, Resets, Limits

2026-10-19cdnping editors4 min read

Alibaba Cloud's Coding Plan (called Bailian in Chinese, now branded Alibaba Cloud Model Studio) currently sells only one tier, Pro, at ¥200/month. Quota is counted in "requests," not tokens. This article is based on the Coding Plan documentation in Alibaba Cloud's help center as it read on October 17, 2026. Prices, quotas, and terms can change at any time, so check the official page before you buy.

#What can you buy right now?

• Pro: ¥200/month. New customers pay ¥39.9 for the first month of their first subscription, then ¥200/month on renewal. The checkout page is the final authority.
• Limited supply: Slots are limited and first come, first served. More are released every day at 9:30 Beijing time.
• Lite: New purchases stopped on March 20, 2026. Renewals and upgrades stopped at 18:00 on April 13. Existing subscribers can keep using it until it expires.
• Models: The documentation lists recommended models from the Qwen, Kimi (Moonshot AI), GLM (Zhipu), and MiniMax families. Names and versions change, so see the official page.

#How is quota counted, and when does it reset?

Pro has three layers of limits. Hitting any one of them restricts you.

The key point is that one question is not one request. According to the documentation, what gets deducted is the number of actual model calls: roughly 5 to 10 for a simple task, and roughly 10 to 30 or more for a complex one, depending on difficulty, context, and tool use. An agent reads files, runs commands, and then edits code, and every round is a call, so usage burns faster than you would expect.

#How many tasks can you do per day?

This is a rough estimate from the official numbers. It is an example, not a promise.

• The monthly cap is 90,000 requests, or about 3,000 per day over 30 days.
• At an average of 10 calls per task, that is about 300 tasks per day. At 30 calls per task, about 100 per day.
• The weekly cap of 45,000 averages about 6,400 per day, so the monthly cap usually becomes the binding constraint first, unless your usage is concentrated on a few days.
• The 5-hour window of 6,000 requests is the layer most likely to be hit first during sustained, heavy agent work. When you hit it, wait for the rolling release.

To find your real consumption, run a fixed task a few times, compare the usage change in the console, and work out the average number of calls per task.

#Dedicated key and endpoints

• The key starts with sk-sp- and is not interchangeable with the pay-as-you-go sk- keys.
• OpenAI-compatible endpoint: https://coding.dashscope.aliyuncs.com/v1
• Anthropic-compatible endpoint: https://coding.dashscope.aliyuncs.com/apps/anthropic

The documentation warns that a general-purpose sk- key cannot be used for Coding Plan, and using one by mistake is billed pay-as-you-go. Always use an sk-sp- key together with an endpoint containing coding.dashscope.aliyuncs.com.

#How do you configure Claude Code?

This official page does not give specific Claude Code variables or model settings, so see Alibaba Cloud's Claude Code integration docs. The general approach is to put the Anthropic-compatible endpoint and the dedicated key into Claude Code's custom endpoint settings. For details, see our posts on configuring custom API endpoints for Claude Code, Codex, and OpenCode and pointing Claude Code at a custom base URL. After configuring, send one simple request, then check the console to confirm it was charged against Coding Plan quota.

#What are the usage restrictions?

1. No refunds.
2. Interactive use inside coding tools only. The documentation prohibits automated scripts, self-built application backends, and non-interactive batch calls. Violations may lead to suspension of the subscription or a banned key. Do not wire it into scheduled jobs or batch pipelines.
3. Personal use only. Sharing is prohibited and may cause your benefits to be restricted.
4. Data. During the subscription, model inputs and generated content may be used for service improvement and model optimization. Stopping use ends further authorization, but data already authorized is not affected. See the Model Studio service agreement for details. Evaluate this clause before you put company code or sensitive content through it.

#A request failed: quota or outage?

Work through this in order.

1. Read the response. Quota limits usually come with a message in the error. The exact fields are defined in the official docs.
2. Compare against the timing. The 5-hour quota recovers on a rolling basis, so a few minutes later released usage may restore access. For the weekly quota, check whether Monday 00:00 has passed. For the monthly quota, check your subscription day.
3. Then check the server side. Open /en/down/qwen and /en/status/dashscope to see whether the official endpoints have had recent probe failures. If several people report errors at once and the endpoint looks unhealthy, it is most likely a server-side problem. If the endpoint is healthy and only you are rejected, check quota and key first.
4. Confirm that you have not mixed up keys and endpoints, as described above.

For how to read status pages, see how to read the OpenAI and Claude status pages. For another plan that also meters by request, see how GLM Coding Plan quota works. It is offered here only as a comparison of quota mechanics, not a ranking. Changes to official prices and terms are also logged at /en/changes.

#Sources

• Alibaba Cloud Model Studio Coding Plan documentation, verified October 17, 2026