First-call checklist before trying a new LLM gateway

I have been using the same small checklist whenever I try a new OpenAI-compatible LLM gateway. It catches most integration failures before I move any real workload.

1. Create one API key and keep it scoped to a test project.
2. Run one tiny chat completion with the cheapest model first.
3. Check whether streaming chunks match the client you already use.
4. Test one tool-call shape if your app depends on tools or agents.
5. Open the usage logs and confirm model name, tokens, timestamp and error mapping are visible.
6. Only then test fallback, higher QPS, or paid traffic.

For Chinese models like Qwen, DeepSeek, GLM and Kimi, the first blocker is often not model quality. It is account setup, payment, model naming and whether the gateway gives enough request logs to debug the first call.

I am collecting small first-call failure cases while building AnLink API notes. If you are testing this path, the beta entry is here: https://anlinkai.com/?ref=r_d0dbd2241021