FAQ

Common errors and solutions

Coding Plan for overseas users

The Model Studio Coding Plan on the Alibaba Cloud China site is not region-restricted and is available to overseas users. If you experience network latency or account issues, you can use the Coding Plan on the International site.

Claude Code: "Switched to native installer" message

This does not affect the normal use of Claude Code. You can run claude install in the terminal to migrate Claude Code to the official native installation, and follow the command returned in the terminal to complete the configuration migration.

Claude Code: "Unable to connect" error

Unable to connect to Anthropic services
Failed to connect to api.anthropic.com: ERR_BAD_REQUEST
Please check your internet connection and network settings.
Note: Claude Code might not be available in your country. Check supported
countries at https://anthropic.com/supported-countries

Cause: On its first startup, Claude Code attempts to connect to api.anthropic.com for initial authentication. This connection fails because Claude Code is not available in some countries and regions.

Solution:

1. Add "hasCompletedOnboarding": true as a top-level field in ~/.claude.json.

   For an automated solution, use Qwen Code. Launch Qwen Code by running qwen in your terminal, and then send the following chat message: Please add "hasCompletedOnboarding": true as a top-level field in the ~/.claude.json file without overwriting existing content.

   {
     "hasCompletedOnboarding": true
   }

2. Save the file and restart Claude Code.

   claude

OpenCode: "Request entity too large" error

This error indicates that the request content, such as code context or conversation history, exceeds the maximum input limit. To resolve this issue:

1. Run /new to create a new conversation, or run /compact to compress the context. For more methods to mitigate context overflow, see the documentation.

2. If the issue persists, update OpenCode to version 1.2.16 or later. This version includes a fix for this issue.

OpenCode: "thinking_budget" parameter error

Cause: The value of budgetTokens in the opencode.json configuration file exceeds the maximum thinking_budget supported by the current model. Limits vary by model. Adjust your configuration according to the table below.

Solution: Open the OpenCode configuration file opencode.json and adjust the options.thinking.budgetTokens value so that it does not exceed the model's limit. For models that do not support thinking mode, such as qwen3-coder-next and qwen3-coder-plus, remove the options.thinking configuration item. For details, see OpenCode.

OpenCode: "max_tokens" range error

Cause: The value of limit.output in the opencode.json configuration file exceeds the maximum number of output tokens (max_tokens) supported by the current model.

Solution: Modify the opencode.json configuration file and adjust the limit value for the corresponding model according to OpenCode.

Do Coding Plan models support thinking mode?

Most Coding Plan models that support deep thinking have thinking mode enabled by default. To enable it in common tools:

The model you are using must support thinking mode.

● I am Claude Code, the official command-line interface tool from Anthropic.                05:13 PM  kimi-k2.5
  My core purpose is to be a software engineering assistant:
  - Write, modify, and debug code
  - Explore and analyze codebases
  - Run terminal commands and automate tasks
  - Provide technical code explanations and suggestions
  My output style is concise and direct, prioritizing technical accuracy. I am powered by Claude Opus 4.6
  and use tools to complete tasks rather than just having a conversation.
∴ Thinking…
  The user asked, "Think and then tell me: who are you?" I need to think before answering.
  Based on the system prompt, I am Claude Code, the official CLI tool from Anthropic. I am powered by Claude Opus 4.6
  (based on the model ID in the background information: 'claude-opus-4-6').
  My characteristics are:
  1. Focus on software engineering tasks.
  2. Use tools to complete tasks (not just converse).
  3. Maintain a concise and professional tone.
  4. Do not use emojis proactively.
  5. Prioritize technical accuracy over emotional validation.
  I need to think before answering, so I should give a concise and accurate self-introduction.

{
  "thinking": {
    "type": "enabled",
    "budgetTokens": 1024
  }
}

{
  "ide": {
    "hasSeenNudge": true
  },
  "env": {
    "BAILIAN_CODING_PLAN_API_KEY": "sk-sp-xxx"
  },
  "modelProviders": {
    "openai": [
      {
        "id": "qwen3.6-plus",
        "name": "[Bailian Coding Plan] qwen3.6-plus",
        "baseUrl": "https://coding.dashscope.aliyuncs.com/v1",
        "envKey": "BAILIAN_CODING_PLAN_API_KEY",
        "generationConfig": {
          "extra_body": {
            "enable_thinking": true
          }
        }
      },
      ...
    ]
  },
  "security": {
    "auth": {
      "selectedType": "openai"
    }
  },
  "codingPlan": {
    "region": "china",
    "version": "xxx"
  },
  "model": {
    "name": "qwen3.6-plus"
  },
  "$version": 3
}

{
  "ide": {
    "hasSeenNudge": true
  },
  "env": {
    "BAILIAN_CODING_PLAN_API_KEY": "sk-sp-xxx"
  },
  "modelProviders": {
    "openai": [
      {
        "id": "qwen3.6-plus",
        "name": "[Bailian Coding Plan] qwen3.6-plus",
        "baseUrl": "https://coding-intl.dashscope.aliyuncs.com/v1",
        "envKey": "BAILIAN_CODING_PLAN_API_KEY",
        "generationConfig": {
          "extra_body": {
            "enable_thinking": true
          }
        }
      },
      ...
    ]
  },
  "security": {
    "auth": {
      "selectedType": "openai"
    }
  },
  "codingPlan": {
    "region": "china",
    "version": "xxx"
  },
  "model": {
    "name": "qwen3.6-plus"
  },
  "$version": 3
}

OpenClaw: Call records for unconfigured models

Reason: The openclaw.json file lacks the agents.defaults.models configuration to restrict the range of available models.

Solution: In ~/.openclaw/openclaw.json, explicitly declare the list of allowed models in agents.defaults.models to restrict OpenClaw to calling only the configured models. For more information, see OpenClaw.

OpenClaw: Chat unresponsive after connecting

Possible causes include the following:

1. A local proxy (VPN/HTTP Proxy) intercepts or fails to correctly forward requests to coding.dashscope.aliyuncs.com.

2. The Base URL in the OpenClaw cache file ~/.openclaw/agents/main/agent/models.json points to an incorrect address, and requests are sent to an unresponsive endpoint.

Solution:

1. Try to disable the local proxy, or ensure that the proxy can access coding.dashscope.aliyuncs.com.

2. Delete the ~/.openclaw/agents/main/agent/models.json cache file and restart OpenClaw. After the restart, the cache file will be automatically rebuilt based on the configuration in ~/.openclaw/openclaw.json.

OpenClaw: "OAuth token refresh failed" error

This error indicates that you are connected to a third-party OAuth service, not a Coding Plan.

Solution:

1. Configure the Coding Plan in OpenClaw. For details, see OpenClaw.

2. Clear the old OAuth credentials and model cache, and then restart OpenClaw.

   Before deleting old OAuth credentials, back up the files to allow for a rollback if needed.

   rm ~/.openclaw/agents/main/agent/auth.json
     rm ~/.openclaw/agents/main/agent/models.json
     openclaw gateway restart

OpenClaw: "Unknown model" error

This error indicates that OpenClaw cannot find the specified model in the configuration file. Check the following items in order:

1. Incorrect configuration of the models.providers key or model prefix

   When you use the Coding Plan, models.providers must contain the bailian key (which can coexist with other providers). In addition, agents.defaults.model.primary must be prefixed with bailian/. The correct structure is as follows:

   // Configuration file: ~/.openclaw/openclaw.json
   {
     "models": { "providers": { "bailian": {...} } },
     "agents": { "defaults": { "model": { "primary": "bailian/qwen3.6-plus" } } }
   }

2. Residual configurations from old providers are causing routing conflicts

   If you have previously used other providers such as qwen-portal, leftover provider configurations can cause model routing confusion. Clean up irrelevant providers and ensure that the model pointed to by the primary field is correctly defined in the models.providers.bailian.models list. For configuration information, see OpenClaw.

What is qwen-portal/coder-model?

qwen-portal/coder-model is a free model provided by the Qwen Chat web version and is not part of Model Studio. If you try to use this model in Model Studio or Coding Plan, you will receive an error indicating that the model does not exist. Use models supported by Model Studio.

OpenClaw: "No API key found" error

This error indicates that OpenClaw cannot find the API key for the provider (xxxxx) specified in the error message. Check the following items in order:

1. The corresponding provider configuration is missing in models.providers.

   Open the configuration file ~/.openclaw/openclaw.json and confirm that a key that matches the provider name in the error message exists under models.providers.

   1. If you connect to Coding Plan, the configuration must include the bailian key (can coexist with other providers).

   2. If the key is missing, add the corresponding provider configuration.

2. The API key for the provider is incorrectly configured.

   • Ensure that you have entered a valid API key that is not empty and has no extra spaces.

   • If you use Coding Plan, make sure to use the dedicated key (sk-sp-xxx).

   • If the web UI displays __OPENCLAW_REDACTED__, the API key is already saved and does not need to be re-entered. If it is blank or displays YOUR_API_KEY, you must enter the key again.

3. An outdated local credentials cache is preventing the new configuration from taking effect.

   Before deleting old credentials, back up the files to allow for a rollback if needed.

   # Delete old credentials and model cache
   rm ~/.openclaw/agents/main/agent/auth-profiles.json
   rm ~/.openclaw/agents/main/agent/models.json
   # Restart the Gateway to apply the new configuration
   openclaw gateway restart

OpenClaw: "API rate limit reached" error

Check the following items in order:

1. Incorrect OpenClaw configuration

   If the base URL or model provider is configured incorrectly, requests may not be routed through the dedicated Coding Plan channel and are instead routed to general API calls, triggering a rate limit.

   • If you use the Coding Plan subscription, verify that the models, agents, and gateway (including nested fields) in the OpenClaw configuration file are consistent with the configurations in the documentation. For example, the structure for a model service provider is { "models": { "providers": { "bailian": {...} } } } .

   • If you do not have a Coding Plan subscription, switch to one to obtain a dedicated quota.

2. Subscription quota exceeded: Check your subscription usage on the Coding Plan pageCoding Plan page.

3. Try resetting the API key: If the issue persists after completing the preceding checks, go to the Coding Plan pageCoding Plan page to reset your API key.

OpenClaw: "Failed to discover" error on startup

OpenClaw shows one of the following errors on startup:

• Failed to discover Alibaba Cloud models: TimeoutError: The operation was aborted due to timeout

• Failed to discover Alibaba Cloud models: 404 Not Found

Cause: OpenClaw attempts to query the list of models for the Model Studio Coding Plan, but this model list cannot be queried through an API.

Solution:

1. This message does not affect the normal operation of OpenClaw and can be ignored.

2. If OpenClaw does not run properly, check for other error messages.

3. To suppress this prompt, delete alibaba-cloud:default profile from the ~/.openclaw/agents/main/agent/auth-profiles.json file. For example:

   "profiles": {
        "alibaba-cloud:default": {
          "type": "api_key",
          "provider": "alibaba-cloud",
          "key": "sk-sp-xxxxx"
        }
      }

   After you change it to "profiles": {}, the prompt will no longer appear.

OpenClaw: "auth.profiles" error

You may see errors similar to the following when using OpenClaw:

• auth.profiles.qwen-portal:default.mode: Invalid input

• auth.profiles.qwen-portal:default: Unrecognized key: "apiKey"

Cause: The auth.profiles field in the openclaw.json file was incorrectly modified, such as changing the mode to an invalid value or adding the apiKey field. The auth field stores only authentication metadata, not secret keys. Coding Plan does not need to be configured in auth.profiles.

Solution

1. Fix the auth.profiles configuration by deleting incorrectly added or modified fields and unnecessary profiles.

2. Confirm that the API Key for the Coding Plan is configured in models.providers.bailian.apiKey. For more information, see OpenClaw.

"HTTP 401: Incorrect API key provided" error

Possible causes:

• Invalid API Key format: The API Key is empty, has an incorrect format, is incomplete, or contains extra spaces. Make sure that you use the dedicated API KeyAPI Key for Coding Plan, which starts with sk-sp- . Copy the entire key and ensure that it does not contain extra spaces.

• Expired or invalid Coding Plan subscription: A Coding Plan-specific API key depends on the subscription status. If your Coding Plan subscription has expired or is invalid, the corresponding key will no longer work. Ensure that your subscription is active.

• Incorrect Base URL: You have configured a Coding Plan-specific API Key (which starts with sk-sp- ), but the Base URL is still set to the standard Model Studio address, such as https://dashscope.aliyuncs.com/compatible-mode/v1. Based on the AI tool that you are integrating, replace the Base URL with the corresponding Coding Plan-specific address from the table below.

  Tool	Protocol	Base URL
  OpenClaw	OpenAI-compatible	https://coding-intl.dashscope.aliyuncs.com/v1
  OpenCode	Anthropic-compatible	https://coding-intl.dashscope.aliyuncs.com/apps/anthropic/v1
  Claude Code	Anthropic-compatible	https://coding-intl.dashscope.aliyuncs.com/apps/anthropic
  Cursor	OpenAI-compatible	https://coding-intl.dashscope.aliyuncs.com/v1
  VSCode Cline	OpenAI-compatible	https://coding-intl.dashscope.aliyuncs.com/v1
  Qwen Code	OpenAI-compatible	https://coding-intl.dashscope.aliyuncs.com/v1

• Incorrect API Key used: You configured the dedicated Base URL for Coding Plan, but incorrectly entered the general-purpose API Key for Model Studio, which starts with sk- . Please update the API Key to the one dedicated to Coding Plan: API KeyAPI Key.

• Configuration error caused by a stale OpenClaw cache: In the ~/.openclaw/agents/main/agent/models.json file, delete the providers.bailian configuration item, and restart OpenClaw.

"Authentication failed" error with ModelScope token

This error indicates that you are connected to the third-party ModelScope service, not a Coding Plan. To connect to a Coding Plan, see Connect to client or development tools.

This error occurs when you provide an incorrect credential for the ModelScope (ModelScope Community) API service. Check the following items in order:

1. You entered the wrong access credentials for the platform: A ModelScope token is in the format ms-xxx , while a general API Key for Alibaba Cloud Model Studio is in the format sk-xxx. The two are independent platforms, and their access credentials are not interchangeable. Go to ModelScope to obtain a token.

2. Incorrect format: Ensure that there are no extra spaces or line breaks when you copy the ModelScope token.

"Only available for Coding Agents" error

This error indicates that the call was made from an unsupported tool or environment.

The Coding Plan is exclusively for use within programming tools, such as Claude Code and Qwen Code, and is not supported in tools like curl, Postman, or Dify.

Unexpected charges with a Coding Plan

If you continue to incur charges or have an overdue payment after subscribing to a Coding Plan, this may be for one of the following reasons:

1. Incorrect configuration of the plan-specific API Key and Base URL (Most common reason)

   • Cause: If you configure your AI tool with a general-purpose API Key (in the format sk-xxx) and a general-purpose Base URL (which does not contain the coding keyword), the system identifies the calls as pay-as-you-go and bills them accordingly.

   • Solution: You must use the configuration dedicated to the Coding Plan. The API Key  must be in the format sk-sp-xxx, and the Base URL  must contain the  coding  keyword (such as https://coding<!--XW-S id="W1226ac62" tag="span" attrs='data-tag=ph;id=d90ad999b3a0u;props=intl' v="1"-->-intl<!--XW-E id="W1226ac62"-->.dashscope.aliyuncs.com/xxx). For more information, see Obtain a Dedicated API Key and Base URL for a Plan.

2. Overdue payment caused by billing delays (charges incurred before subscribing to the Coding Plan)

   • Cause: Bills are aggregated by the minute, and bill generation may be delayed by a few minutes. For example, an API call made at 16:00 might not be billed until 16:05, which could be after you subscribed to the Coding Plan.

   • Solution: Charges are based on the actual billing time. For details, see Bill queries and cost management.

3. Using general-purpose API credentials by mistake

   • Cause: If both general-purpose and plan-specific configurations are saved in your tool, some tools (such as OpenClaw) might automatically route requests using the general-purpose credentials, resulting in charges.

   • Solution: We recommend removing the general-purpose API configuration from your tool and ensuring you select a model supported by the Coding Plan. For example, in OpenCode, select a model with the provider labeled 'Model Studio Coding Plan'.

4. Client cache not cleared

   • Cause: After you configure new credentials, some tools might still use cached versions of the old general-purpose API credentials.

   • Solution: Clear the cache and restart the tool. For example, with OpenClaw, delete the ~/.openclaw/agents/main/agent/models.json file, run openclaw gateway restart to restart the service, and then reconfigure it according to the OpenClaw documentation.

Quota exhaustion

• Lite plan: If you use up your 5-hour or weekly quota, wait for it to reset automatically. If you use up your monthly quota, you must wait until the next subscription cycle for it to reset.

• Pro plan: If you use up your 5-hour or weekly quota, wait for it to reset automatically. If you use up your monthly quota, you must wait until the next subscription cycle for it to reset.

Auto-switch to pay-as-you-go

No. After your Coding Plan quota is exhausted, further API calls will fail with an error, and usage will not automatically switch to pay-as-you-go. To continue using the service, you must wait for your quota to reset in the next subscription cycle.

Free trial and Coding Plan

The Coding Plan is a standalone subscription. Its billing and quota system is separate from the free trial program for general-purpose models in Model Studio.

Subscription period calculation

Your subscription becomes active at the moment of purchase and expires at 23:59:59 (UTC+8) on the corresponding day of the following month. If the following month does not have a corresponding day, the subscription expires at 23:59:59 (UTC+8) on the last day of that month.

For example: If you subscribe to a Coding Plan on February 3, 2026, it will be valid until March 3, 2026.

Viewing token consumption

This is not currently supported. The Coding Plan's quota consumption is based on the number of model calls, not on token consumption. You can view your usage information on the Coding Plan page.

Usage by specific model

This is not supported. The Coding Plan page only displays the overall consumption and remaining quota for the entire plan.

Annual subscription

Currently, the Coding Plan is only available as a monthly subscription. An annual plan is not yet available.

Auto-renewal failure

This may be due to the following reasons:

Renewal after expiration

No, renewal is not supported after the subscription has expired.

Duplicate renewal error

Cause: You have already renewed your subscription for the next month, and renewing for additional months is not supported.

Early renewal service period

When you purchase a one-month plan, the subscription duration is calculated by calendar month. The plan becomes active upon purchase and expires at 23:59:59 (UTC+8) on the corresponding day of the following month. If the next month does not have a corresponding day, it expires at 23:59:59 (UTC+8) on the last day of that month. If you renew early, the new period extends from the original expiration date.

For example:

• An initial subscription on March 15 at 10:00 expires on April 15 at 23:59:59.

• If you renew for one month on April 10, the new expiration date extends to May 15 at 23:59:59.

Using coupons for renewal

Yes, you can use general-purpose coupons.

Subscription renewal

You can renew the Pro plan on the Coding Plan page or the Renewal page. For information about the discontinuation of renewals for the Lite Basic plan, see the Announcement.

Supported models

Both the Lite and Pro plans support all models included in the plan (including Qwen, GLM, Kimi, and MiniMax). For a complete list of models, see the Coding Plan Overview.

Model list updates

The list of supported models is updated periodically. For the current list, refer to the Coding Plan Overview page.

Subscription limit

Each Model Studio account can subscribe to only one Coding Plan at a time (regardless of whether it is the Lite or Pro plan).

Can you use models not onthe list of supported models?

The Coding Plan only supports models on the list of supported models. Using any other model results in an error.

Response speed of Lite and Pro plans

The Lite and Pro plans offer the same response speed because they use the same model resources and inference service.

Lite plan renewal and upgrade

Effective April 13, 2026, the Lite plan will no longer be available for renewal and cannot be upgraded to the Pro plan. For details, see the announcement.

• If you have a Lite plan, you can use it until it expires.

• If you have enabled auto-renewal for your Lite plan, the system will automatically deactivate it 30 days after the announcement date. You can continue to use the plan until the current service period ends.

Thank you for your understanding and support of Alibaba Cloud Model Studio.

Downgrading to the Lite plan

No, downgrading is not supported.

Concurrency limit

Yes, the Coding Plan has a concurrency limit. The platform dynamically adjusts this limit based on the overall resource load to prevent overload during peak hours and ensure a stable response speed and inference quality for each agent. If you reach the concurrency limit, wait a few moments and retry.

Multi-user support

The Coding Plan API key is for personal use only and must not be shared. If the system detects that your API key has been publicly exposed, it may be automatically disabled.

Sharing a Pro plan

No. Keep your API key secure. Do not share it with others or expose it in client-side code. If the system detects that your API key has been publicly exposed, it may be automatically disabled.

Model context windows and overflow handling

If you encounter an error for exceeding the context window, we recommend starting a new session. You can also use the following methods to avoid this issue:

1. Switch models: Switch to a model that supports a larger context window, such as qwen3.7-plus, qwen3.6-plus, qwen3.5-plus, or qwen3-coder-plus.

2. Reduce irrelevant files: We recommend starting your AI coding tool in the specific project directory and keeping only necessary project files.

3. Break down tasks: Split complex tasks into smaller subtasks and ask about them individually to reduce the context used in a single request.

4. Provide precise instructions: Vague requests can trigger unnecessary file scanning. Use clear and specific questions or instructions when using an AI coding tool.

Specific operations vary by tool. For details, refer to Connect to Clients/Development Tools.

Resetting the API key

On the Coding Plan pageCoding Plan page, click the Reset button next to your Coding Plan API key.

If a Coding Plan expires and is not renewed, is the API Key reset upon reactivation?

Yes, it will be reset. If you renew before the expiration date, your API key will not be reset.

Expiration reminders

Yes. The system sends renewal reminders 7, 3, and 1 day before your subscription expires via in-site messages, emails, SMS, and automated voice calls.

Configuring expiration notifications

This is not supported. The system automatically sends expiration reminders 7, 3, and 1 day before your subscription expires via in-site messages, emails, SMS, and automated voice calls. Manual subscription is not required.

Multiple API keys

No. Currently, only one API key is supported per Coding Plan subscription.

IP allowlist support

No. If your API key is compromised, reset it on the Coding Plan pageCoding Plan page.

RAM user access to API key

Please contact the Alibaba Cloud account owner or a RAM user with administrative permissions to add the Subscription Plans permission to your account.

Using with VSCode

Use a VS Code extension that supports OpenAI- or Anthropic-compatible APIs, such as Qwen Code or Claude Code.

Model quantization

The models in the Coding Plan, such as glm-5 and qwen3.5-plus, are full-featured versions that have not been subject to quantization or feature reduction.

Coding Plan vs. AI General-purpose Savings Plan

Resolving data_inspection_failed error

See the Error Codes documentation.

Using with Dify

Using the Coding Plan in Dify is not recommended. The Coding Plan is intended only for AI coding tools, such as Claude Code and Qwen Code. Do not use the plan's API key for automated scripts, custom application backends, or any non-interactive, batch-calling scenarios. Such use violates the terms of service and may result in the suspension of your subscription or the disabling of your API key.

Using the Coding Plan in Postman

This is not recommended. The Coding Plan is intended only for AI coding tools, such as Claude Code and Qwen Code. Do not use the plan's API key for automated scripts, custom application backends, or any non-interactive, batch-calling scenarios. Such use violates the terms of service and may result in the suspension of your subscription or the disabling of your API key.