AI Performance & Quality
Models Comparison
Choose the model behavior that best fits your support use case.
Choose a model for your workflow
Select a model from the current model picker in Instructions or Settings. Model availability and limits depend on the deployment and plan; the picker is the current source of available choices.
Compare on the same questions
| Criterion | Check |
|---|---|
| Accuracy | Does the answer match the source and avoid invented details? |
| Actions | Does the model select the correct action and collect the required inputs? |
| Latency | Measure the complete customer-visible response. |
| Credits | Compare actual usage for the same set of tasks. |
| Style | Check tone, length, and language against your instructions. |
Run a comparison
- Use a representative set of product questions, follow-ups, unknowns, and action requests.
- Run each set in fresh conversations with the same sources and instructions.
- Change the model and repeat the same scenarios.
- Compare outcomes and usage before selecting the model for production.
Token limits and instructions
Keep the response token limit large enough for the expected answer or tool result. A more capable model cannot supply a missing policy or repair a disconnected action. Correct the source or workflow first.