Choosing a model

Available models

The Model dropdown in an agent's Settings tab sets both the model and the provider together — you don't configure them separately:

• Claude Haiku 4.5 — the default and recommended choice for most agents. Fast and inexpensive, and grounded well by retrieval, which is where most of an agent's answer quality actually comes from.
• Claude Sonnet 5 — a premium tier for agents doing more complex reasoning, multi-step tool use, or nuanced writing.
• GPT-4o mini, GPT-4o, GPT-4.1 mini, GPT-4.1 — OpenAI alternatives, also selectable per agent.
• External agent — Microsoft Azure AI Foundry — delegates the agent entirely to an agent you've built in Azure AI Foundry, instead of running on a hosted model here. See Creating & configuring an agent for the connection fields, and set up the underlying Azure connection under AI providers first.

There's no bring-your-own-API-key option — every model runs on the shared platform key, so switching models is just a dropdown change with no credentials to manage.

Picking a model

Start with the default (Haiku 4.5) for new agents. Move to Sonnet 5 or a GPT‑4.x tier if you notice the agent struggling with multi-step tool sequences or complex synthesis across several knowledge sources — not as a first fix for wrong answers. If answers are wrong or missing, the far more common cause is a gap in your knowledge base or retrieval settings that are too strict/loose, not the model itself. Check Retrieval & grounded answers before upgrading the model.

Temperature

Temperature (0–1.5, in Settings) controls variation in wording and phrasing, not factual accuracy — grounding and retrieval control accuracy. Keep it low (the default range) for support agents where consistent, predictable answers matter more than creative phrasing.