Documentation
Reasoning Models in the NexoRouter documentation.
Reasoning Models
Status: Use model-specific testing.
Reasoning models can be slower and more expensive than standard chat models. They also vary in how clients expose reasoning controls.
First test
Use a small prompt and a longer timeout:
Explain the tradeoffs of retrying a failed API request in three bullets.Client timeout
Start with at least 60 seconds for ordinary chat models. For slower reasoning-style models, allow up to 180 seconds before adding retries.
Cost control
- Use a key budget for experiments.
- Keep prompts short.
- Avoid running agent loops until the text response path is verified.
- Watch Usage Logs for token usage and cost.