Model Selection
Ce contenu n’est pas encore disponible dans votre langue.
SynapsAI supports multiple large language models (LLMs). The model you choose affects response quality, speed, cost, and the types of tasks your agent can handle.
How to select a model
Section titled “How to select a model”- Select your agent from the Agents list.
- Go to Agent Settings.
- Find the Model dropdown.
- Select the model you want to use.
- Click Save.
Changes take effect immediately for new conversations.
Choosing the right model
Section titled “Choosing the right model”Different models excel at different tasks. Consider these factors:
| Factor | What to consider |
|---|---|
| Response quality | More capable models produce more accurate, nuanced answers — but cost more per message. |
| Speed | Smaller models respond faster. If latency matters (e.g., live chat), consider a faster model. |
| Cost | Each model has a different cost per token. Higher-tier models consume more credits per message. |
| Context window | Models with larger context windows can process more conversation history and knowledge at once. |
| Language support | Some models handle non-English languages better than others. |
When to use a more capable model
Section titled “When to use a more capable model”- Complex questions that require reasoning across multiple documents.
- Nuanced conversations where tone, empathy, and context matter.
- Multilingual support where translation quality is critical.
- Sensitive topics where accuracy is more important than speed.
When a lighter model is sufficient
Section titled “When a lighter model is sufficient”- Simple FAQ lookups where the answer is a direct match from the knowledge base.
- High-volume, low-complexity chats where speed and cost matter most.
- Internal tools where response polish is less important.
Testing models
Section titled “Testing models”Use the Playground to compare models side by side:
- Set a model in Agent Settings.
- Ask a set of representative questions in the Playground.
- Note the response quality, speed, and accuracy.
- Switch to a different model and repeat.
- Choose the best balance for your use case.
Credits and usage
Section titled “Credits and usage”Each message consumes credits based on the model used and the number of tokens processed (input + output). More capable models use more credits per message.
Monitor your credit usage in Plans & Billing.
Next steps
Section titled “Next steps”- Agent Settings — configure temperature and token limits for your selected model.
- Configure Behavior — write a system prompt optimized for your chosen model.
- Playground — test model performance with real questions.