Troubleshooting
Background
Due to network conditions, vendor service instability, and the inherent characteristics of large language models, you may encounter various issues during usage, including but not limited to: timeouts, request failures, and unstable responses.
If you have requirements around success rates, custom timeout durations, or other model invocation strategies, you can find a suitable solution in the cases below.
All solutions below are engineering optimization measures designed to improve request success rates and response stability. They cannot fundamentally resolve instability issues on the model vendor's side.
If your project already has mature engineering strategies in place (such as custom retry logic, circuit breakers, or fallback handling), it is recommended to use your own approach first.
Case List
- Timeouts and Model Instability
- Issue Diagnosis and Tracing
- Gemini 429 Rate Limiting and Provisioned Throughput
- Employee Verification Failure (2850 / 2851)
Minor issues specific to certain APIs (Gemini 3 multi-turn conversation 400 errors, Embedding dimensions, Realtime connections, and video status polling) have been merged into their respective guide pages: see Gemini Thought Signatures, Embeddings, Realtime, and Video Generation.
Getting Help
If the above content does not resolve your issue, please contact technical support via Get Help.