Model Studio cookbook

You can already send one request. These are problem-shaped. Activation is in the tutorial.

Move existing OpenAI code

Official rule: change the API key, base URL, and model name. Intro

1. Read process.env.DASHSCOPE_API_KEY.
2. Point baseURL at that region's OpenAI-compatible host and fill {WorkspaceId} (Beijing / Singapore / Tokyo / Frankfurt). US Virginia follows the model list.
3. Use an ID from the list. gpt-4o will not resolve.
4. Streaming and tools exist only if that model lists them. Compatible ≠ feature-parity.

Copy Node / Python / curl from First API call.

Do not mix the three credential pairs

This is the usual source of invalid_api_key and surprise invoices. Coding Plan, Token Plan

Billing	Key looks like	Where the base URL comes from
Pay-as-you-go	sk- or sk-ws	First API call / Models
Coding Plan	sk-sp- from the plan page	OpenAI https://coding.dashscope.aliyuncs.com/v1; Anthropic https://coding.dashscope.aliyuncs.com/apps/anthropic
Token Plan	Dedicated key (official prefix sk-sp-)	Whatever the Token Plan console shows that day — not the Coding Plan host

Official failure modes:

• PAYG key + plan host → invalid_api_key.
• PAYG key + PAYG host → no plan credit, PAYG bill.
• You bought Coding Plan and still see arrears: the tool still has sk- and dashscope.aliyuncs.com.

Coding Plan is coding tools only (official examples: Claude Code, Qoder, Qoder CN, OpenClaw). Using the plan key as your app backend or for batch jobs is a policy violation.

Token Plan requires a real OpenAI or Anthropic protocol. A look-alike UI that does not speak those APIs cannot burn plan credits — use a PAYG sk- / sk-ws key instead.

When filling a client, open the official page for that client. Do not reconstruct screenshots here:

• Index: Clients and developer tools
• Examples: Claude Code, Chatbox

The full Qoder CN / Lingma IDE tutorial is #84. One official fact belongs here: personal Community / Pro can attach Model Studio; Enterprise cannot. Qoder CN

Stop PAYG after the new-user quota

China-site free quota:

1. Beijing only. Valid 90 days from the later of activation, model launch, or approval.
2. Realtime inference only. Not Batch, fine-tune, deploy, or custom models.
3. Each model (including dated snapshots) has its own pool. The page says a model is "usually" 1 million tokens — believe the console row, not the marketing "70 million+" headline.
4. Verified accounts: turn on Free quota only. Exhaustion returns AllocationQuota.FreeTierOnly.
5. Unverified accounts already have that switch forced on.
6. An overdue account cannot call any model, even one that still has free tokens.

The official page advises against Free-quota-only in production (the service just stops). Use spend alerts and key deletion there.

Token Plan / Coding Plan dedicated keys do not consume this free-tier pool.

No-code knowledge-base Q&A

1. Open the console and confirm the region.
2. Follow Build a Q&A app without code.
3. Prefer Agent 2.0 when you have no 1.0 dependency. No in-place upgrade from 1.0.
4. Publish before you call the app API or channels.
5. Knowledge base billing is separate from model inference and does not use savings plans or resource packs. Intro

Fine-tune and dedicated deploy stay on the official pages: training, deploy.

Install the Model Studio CLI

This is Model Studio's own CLI (multimodal tools for local agents), not the Alibaba Cloud OpenAPI plugin aliyun bailian.

Official install: bailian.aliyun.com/cli/install.md

# Node.js ≥ 22.12.0 — npm only
npm install -g bailian-cli
bl --version

# Browser login (preferred)
bl auth login --console

# Smoke test
bl auth status --output json
bl text chat --message "ping" --non-interactive --output json

Commands are bl and bailian. Official skills install: npx skills add modelstudioai/cli --all -g.

Headless / SSH: bl auth login --api-key <pasted-key>. Do not write the key into the repo or the chat log.

Region flag: --region cn|us|intl, default cn.