Check DashScope responses and audio compatibility without guessing.
The fastest way to debug a DashScope integration is to separate three questions: does the native provider work, does the OpenAI-compatible relay preserve the same shape, and does the audio path need a different contract from text responses?
Working rule: treat text, responses, tools and audio as separate checks. A route that works for plain chat does not automatically prove responses or audio compatibility.
1. Decide which path you are testing
Best when you already know the provider-specific SDK and want official behavior.
Best when your app expects the OpenAI surface and you need a drop-in endpoint.
Best treated as its own route, because input and output handling often differ from text.
2. Run the smallest test first
Do one plain text request, then one responses-shaped request, then one audio-related request. If you skip straight to tools or streaming, you lose the cleanest signal.
curl https://api.example.com/v1/responses \
-H "Content-Type: application/json" \
-H "Authorization: Bearer sk-your-token" \
-d '{
"model": "dashscope-route",
"input": "Say hello in one sentence."
}'If the text call works but responses or audio fail, the problem is probably not authentication. It is usually schema shape, routing, or a provider feature gap.
3. Check the response shape before blaming the model
| Check | What to verify | What a failure usually means |
|---|---|---|
| Text response | One small prompt returns a clean answer. | Token, host or base URL mismatch. |
| Responses shape | The relay preserves the fields your client expects. | Adapter transforms are dropping or renaming fields. |
| Tool schema | Required tool fields survive the round trip. | The provider or relay is rewriting the request body. |
| Audio route | Input and output media requirements are handled separately. | Audio is being treated like a normal chat completion. |
4. Keep the adapter honest
When you build or review a relay, keep provider-specific logic behind capability checks. That keeps normal text routes clean while still preserving the special fields that certain DashScope flows expect.
- Log the final outbound host and model id.
- Keep tool payloads stable across round trips.
- Test audio and responses separately from basic text.
- Do not assume one successful chat test proves media support.
5. When a gateway helps
If your main blocker is not the model itself but getting a route that your team can test quickly from outside China, Black Eagle AI can provide a prepaid OpenAI-compatible path while you compare native and relay behavior.
For teams comparing provider behavior, start with one small request and one clean log. That is enough to tell whether the mismatch lives in the upstream provider, the relay, or the client.
Want a small test route for comparison?
Use the request form if you need a prepaid Chinese-model route for your first comparison test, or open the integrations page if you want a general OpenAI-compatible setup guide.
Request access