Qwen models listed as working but fail with Bad_Request / Internal error

[tektual]

Description

Qwen models (e.g. qwen-3.7-plus, qwen-3.5-plus, qwen-plus) are listed as working in the working_results.txt via AnyProvider, but when we pull the same Docker image (hlohaus789/g4f:latest) and test them, every request fails with a Bad_Request / Internal error from the Qwen provider.

Models listed as working

From working/working_results.txt:

AnyProvider|qwen-3.7-plus|text
AnyProvider|qwen-3.7-max|text
AnyProvider|qwen-3.6-plus|text
AnyProvider|qwen-3.5-plus|text
AnyProvider|qwen-plus|text
AnyProvider|qwen-3.5-flash|text
AnyProvider|qwen-3.5-omni-plus|text
AnyProvider|qwen-3.6-max|text
AnyProvider|qwen-3.5-omni-flash|text
AnyProvider|qwen-3-omni-flash|text
AnyProvider|qwen-3-max|text
AnyProvider|qwen-3-vl-plus|text
AnyProvider|qwen-3-coder-plus|text
AnyProvider|qwen-3.5-397b-a17b|text

What we get instead

All Qwen models fail with the same error:

RetryProviderError: RetryProvider failed:
PuterJS: MissingAuthError: API key is required for Puter.js API
Qwen: RuntimeError: Response: {'success': False, 'request_id': '...', 'data': {'code': 'Bad_Request', 'details': 'Internal error...'}}

The AnyProvider fallback only routes to PuterJS (needs auth) and Qwen (returns Bad_Request). No working free fallback exists for Qwen models.

Steps to Reproduce

docker run -d -p 8080:8080 hlohaus789/g4f:latest
curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen-3.7-plus","messages":[{"role":"user","content":"Say hi"}],"max_tokens":20}'

Working free models on the same instance

For comparison, these work without auth:

• gpt-4o ✅
• gpt-4o-mini ✅
• gpt-4.1-nano ✅
• o3-mini ✅
• mistral-nemo ✅

Environment

• g4f version: v7.8.6 (latest Docker image hlohaus789/g4f:latest)
• Container OS: Ubuntu with Python 3.14
• Tested with and without proxy (HTTP_PROXY/HTTPS_PROXY set)

Question

Is it possible the g4f-working CI pipeline is getting false positives for Qwen models? The test response for qwen-3.7-plus just shows Yes — is that coming from a different provider than the Qwen provider that actually scrapes chat.qwen.ai?