The Qwen Coding Plan is mostly closed: the Lite tier stopped taking new buyers on 20 March 2026, and the Pro tier is sold in limited slots that Alibaba restocks daily. Alibaba now points new users to Token Plan. Both are subscriptions for coding agents on Alibaba Cloud Model Studio, and neither is a way to call the API from your own scripts.
What is the state today?
- Coding Plan Lite: closed to new subscriptions since 20 March 2026. Renewals and upgrades ended on 13 April 2026. Existing subscribers can use it until the period they paid for ends.
- Coding Plan Pro: limited slots, first come first served, restocked daily at 00:00 UTC+8. Lite and Pro support the same models and run at the same speed.
- Token Plan: open, in Personal and Team editions. Billed in Credits.
- One Coding Plan per account. A Model Studio account can hold only one at a time.
Alibaba’s pages describe Coding Plan and Token Plan as separate products with separate keys. No page describes moving a Coding Plan subscription into Token Plan, so treat it as a new purchase. Prices are on the Token Plan and Coding Plan pages.
What do the limits look like?
Coding Plan counts model calls, not tokens. A single prompt in an agent can make many calls, depending on the task. Three caps apply at the same time:
- a rolling window of a few hours,
- a weekly quota that resets on Monday at 00:00 UTC+8,
- a monthly quota that resets on your renewal date.
Whichever you reach first pauses your calls. Alibaba says the caps are not cumulative and don’t guarantee even capacity over time. Heavy load can queue or throttle requests.
Token Plan works differently. Each seat gets a fixed monthly allowance of Credits by tier (Standard, Advanced or Premium on Team). Unused Credits don’t roll over. When a seat runs out, API calls are blocked and no pay-as-you-go charge kicks in; you wait for the reset, buy a shared pack that members share for a month, or add seats. Tool calls that use Alibaba’s own tools, such as web search and code interpreter, spend Credits too. A concurrency limit moves with platform load.
Which agents work?
| Agent | Coding Plan | Token Plan |
|---|---|---|
| Qwen Code | Yes | Yes |
| Claude Code | Yes | Yes |
| Codex CLI | Yes, with an old version | Yes |
| OpenCode | Yes | Yes |
| Cline, Kilo CLI | Yes | Through the generic guide |
| Cursor | Yes | Not listed |
| Qoder, WorkBuddy, OpenClaw, Hermes Agent | Yes | Yes |
| Trae, DeepSeek Harness, QwenWork | Not listed | Yes |
Both pages say the list isn’t complete and add a generic guide for other tools. Anything that takes an OpenAI-compatible or Anthropic-compatible base URL will connect, but “will connect” is not “is allowed”: Coding Plan is only for interactive use in programming tools, and Alibaba says using its key in scripts, backends or other non-interactive jobs can get it suspended or revoked. It is for one user and can’t be shared.
Which models are in each plan?
Alibaba’s own Qwen Code setup lists these as examples, and says they aren’t exhaustive.
- Coding Plan: Qwen3.7 Plus, Qwen3.6 Plus, Qwen3.5 Plus, Kimi K2.5, GLM-5 and MiniMax M2.5. Only these exact versions are supported.
- Token Plan Personal: Qwen3.8 Max and Flash, Qwen3.7 Max and Plus, DeepSeek V4.1 Flash and GLM-5.3 and 5.2. Team adds more, including Kimi K2.7 Code.
So Token Plan reaches newer Qwen models than Coding Plan does. Which to use for what is in best Qwen model for coding.
How do I set it up in each agent?
The key and the URL must come from the same plan. Plan keys start with sk-sp-.
| Plan | Key variable in Qwen Code | OpenAI-compatible URL | Anthropic-compatible URL |
|---|---|---|---|
| Token Plan | BAILIAN_TOKEN_PLAN_API_KEY | https://token-plan.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1 | https://token-plan.ap-southeast-1.maas.aliyuncs.com/apps/anthropic |
| Coding Plan | BAILIAN_CODING_PLAN_API_KEY | https://coding-intl.dashscope.aliyuncs.com/v1 | https://coding-intl.dashscope.aliyuncs.com/apps/anthropic |
Qwen Code: run qwen, choose Alibaba ModelStudio (or /auth inside a session), pick Coding Plan or Token Plan, choose the Singapore region and paste the key. Check with qwen -p "Hello".
Claude Code: Anthropic doesn’t support non-Claude models in Claude Code; Alibaba documents the setup. For Coding Plan, every model variable is qwen3.7-plus. Windows PowerShell:
$env:ANTHROPIC_BASE_URL="https://coding-intl.dashscope.aliyuncs.com/apps/anthropic"
$env:ANTHROPIC_AUTH_TOKEN="<your Coding Plan key>"
$env:ANTHROPIC_MODEL="qwen3.7-plus"
$env:ANTHROPIC_DEFAULT_HAIKU_MODEL="qwen3.7-plus"
$env:ANTHROPIC_DEFAULT_SONNET_MODEL="qwen3.7-plus"
$env:ANTHROPIC_DEFAULT_OPUS_MODEL="qwen3.7-plus"
$env:CLAUDE_CODE_SUBAGENT_MODEL="qwen3.7-plus"
claudemacOS or Linux: the same names with export NAME=value. The Token Plan values are on the Qwen page.
Codex CLI: Token Plan uses the config on the Qwen page with Codex 0.94.0 or later. For Coding Plan, Alibaba’s page says Chat Completions and Codex 0.93.0 or earlier:
npm install -g @openai/codex@0.93.0model_provider = "Coding_Plan"
model = "qwen3.7-plus"
[model_providers.Coding_Plan]
name = "Coding_Plan"
base_url = "https://coding-intl.dashscope.aliyuncs.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "chat"Put that in ~/.codex/config.toml, set OPENAI_API_KEY to the Coding Plan key and check with codex exec "Hello". Alibaba’s own quick-start block shows responses for this plan and contradicts its text, so chat on 0.93.0 is the documented pairing.
OpenCode: in ~/.config/opencode/opencode.json, add a provider with "npm": "@ai-sdk/anthropic", the key name bailian-coding-plan, options.baseURL set to https://coding-intl.dashscope.aliyuncs.com/apps/anthropic/v1 and your models under models. The full shape is the Token Plan block on the Qwen page with that URL and key.
Cline and Kilo Code: provider OpenAI Compatible, the Coding Plan OpenAI URL, your key, and a model name such as qwen3.7-plus.
What goes wrong?
- 401: the key is a general
sk-key rather than the plan key, the subscription expired, or a space got into the key. - 403 invalid api-key: a plan key sent to the general Model Studio URL.
- 429: too many calls in a short time. Wait a minute, run fewer agents in parallel, and don’t share the key.
The one-key trap: on Team Edition, a seat’s key is created for that member, and a Personal key doesn’t work as a Team key or the other way round. The platform tells them apart by the key.
Contenox
Contenox takes either plan through its Alibaba Cloud Model Studio provider: paste the key, and for a Coding Plan key set Base URL under Advanced to the plan’s endpoint from the table above. The Model chip then lists the plan’s models. Setup: Qwen coding agent.
Related guides
- Alibaba Cloud Model Studio for coding agents: the endpoints and key rules
- Qwen coding agent: complete configs
- AI coding subscription comparison: how other plans compare
- AI coding costs: why agents burn through quotas