Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

Qwen vision models return 400 "Unexpected item type in content" from DashScope when the image tool is invoked.

Steps to reproduce

1. Configure a Qwen provider with DashScope API (qwen3.7-max or qwen3.7-plus, text+image enabled)
2. Use the image tool with any prompt and any image
3. The request fails with a 400 error

Expected behavior

The image tool should analyze the image and return a description, since both qwen3.7-max and qwen3.7-plus list "text+image" as supported input.

Actual behavior

The image tool fails immediately. The gateway log shows:

model_fallback_decision: candidate_failed reason=format status=400
providerRuntimeFailureKind: schema
errorPreview: "400 InternalError.Algo.InvalidParameter: The provided messages input is invalid. The error info is [Unexpected item type in content.]"

No reply is returned to the user; the fallback chain is exhausted with outcome=chain_exhausted.

OpenClaw version

2026.6.1 (2e08f0f)

Operating system

Ubuntu 26.04

Install method

npm global

Model

qwen/qwen3.7-max

Provider / routing chain

openclaw -> dashscope (api-key)

Additional provider/model setup details

Default provider configured in ~/.openclaw/openclaw.json under models.providers.qwen with api: "openai-completions" and baseUrl pointing to DashScope. The Qwen provider works normally for text-only requests; only multimodal (image) requests fail.

Logs, screenshots, and evidence

[context-overflow-precheck] ... provider=local-llama/qwen25-7b ...
[model-fallback/decision] candidate_failed reason=format status=400
  errorPreview: "400 InternalError.Algo.InvalidParameter: The provided messages input is invalid. The error info is [Unexpected item type in content.]"
  providerRuntimeFailureKind: schema
  fallbackStepFinalOutcome: chain_exhausted
[media-understanding] image: failed (0/1) reason=Image model failed (qwen/qwen3.7-max)

Impact and severity

• Affected: all users relying on Qwen via DashScope for image understanding
• Severity: High (blocks image analysis entirely)
• Frequency: 100% — every image tool call fails
• Consequence: image tool is completely unusable with Qwen/DashScope providers

Additional information

• Text-only requests to the same Qwen provider work without issues.
• WebChat image recognition fails when using coding.dashscope base URL for Qwen models #67759 and Qwen: qwen3.6-plus image understanding blocked on Coding Plan endpoint despite model supporting vision #63654 were previous Qwen vision fixes but addressed endpoint/model routing, not content message format.
• The error reason "format" and "schema" suggests the image content block is structured in a way DashScope rejects.