Summary

qwen3.6-plus is a multimodal model that supports image understanding, but the bundled qwen provider explicitly suppresses it when the configured endpoint is the Qwen Coding Plan (coding.dashscope.aliyuncs.com). This prevents the image tool and media-understanding pipeline from using qwen3.6-plus for vision tasks, even though the model itself supports it.

Why it matters

• Users on the Coding Plan subscription cannot use qwen3.6-plus for image analysis via the image tool.
• The model does support image understanding on Alibaba DashScope, but OpenClaw treats it as unavailable on the Coding Plan endpoint.
• This forces users to either switch to the Standard (pay-as-you-go) endpoint or use a different model for vision, even when their Coding Plan key should theoretically work.

What changed

The restriction is implemented in extensions/qwen/index.js:

// Line ~138-155
normalizeConfig: ({ providerConfig }) => {
  if (!isQwenCodingPlanBaseUrl(providerConfig.baseUrl)) return;
  const models = providerConfig.models?.filter((model) => model.id !== QWEN_36_PLUS_MODEL_ID);
  return models && models.length !== providerConfig.models?.length ? {
    ...providerConfig,
    models
  } : void 0;
},
suppressBuiltInModel: (ctx) => {
  // ...
  return {
    suppress: true,
    errorMessage: "Unknown model: qwen/qwen3.6-plus. qwen3.6-plus is not supported on the Qwen Coding Plan endpoint; use a Standard pay-as-you-go Qwen endpoint or choose qwen/qwen3.5-plus."
  };
}

This causes the image tool to fail with:

Unknown model: dashscope/qwen3.6-plus

Repro + Verification

Environment

• OS: Linux (Debian)
• Runtime: OpenClaw v2026.4.9 (npm global install)
• Provider: Qwen (bundled)
• Endpoint: https://coding.dashscope.aliyuncs.com/v1
• Model: qwen3.6-plus
• Config: custom dashscope provider with openai-completions API style

Steps

1. Configure a Qwen provider pointing to the Coding Plan endpoint:

   {
     "models": {
       "providers": {
         "dashscope": {
           "baseUrl": "https://coding.dashscope.aliyuncs.com/v1",
           "apiKey": "...",
           "api": "openai-completions",
           "models": [{ "id": "qwen3.6-plus", "input": ["text", "image"] }]
         }
       }
     },
     "agents": {
       "defaults": {
         "imageModel": { "primary": "dashscope/qwen3.6-plus" }
       }
     }
   }

2. Send an image via Feishu (or any channel) and ask the model to analyze it.

3. Observe that the image tool fails with Unknown model: dashscope/qwen3.6-plus.

Expected

• qwen3.6-plus should be usable for image understanding when the Coding Plan endpoint supports it.
• The image tool should successfully call the model with the image attachment.

Actual

• The image tool throws Unknown model: dashscope/qwen3.6-plus.
• The model hallucinates image content instead of actually analyzing it.

Evidence

• Gateway log:

  [tools] image failed: Unknown model: dashscope/qwen3.6-plus

• Session transcript shows the model receiving [media attached: /path/to/image.jpg] but never invoking the image tool successfully.
• The restriction is explicitly coded in extensions/qwen/index.js (lines ~138-155).

Compatibility / Migration

• This is not a regression; it's a deliberate design choice in the current release.
• However, it creates a capability gap for Coding Plan users who expect qwen3.6-plus to handle vision tasks.

Suggested fix

1. If Alibaba officially supports qwen3.6-plus image understanding on the Coding Plan endpoint, remove the suppression logic.
2. If support is uncertain, consider making this behavior configurable (e.g., allowCodingPlanVision: true) rather than hard-blocking it.
3. Update the documentation to clarify whether qwen3.6-plus vision is officially unsupported on Coding Plan, or if this is a conservative compatibility guard that could be relaxed.

References

• Official docs state: qwen3.6-plus input includes text, image with 1,000,000 context.
• Docs also note: "Prefer Standard endpoints when you need this model" and "Coding Plan support can lag behind the public catalog."
• This issue is about whether the lag is still current, or if the restriction can now be safely lifted.