Skip to content

Choose an LLM Model

Open Chat Studio supports a range of AI models from providers such as OpenAI, Anthropic, Google, and others. Each model has different strengths. This guide helps you pick the right one for your chatbot without needing a deep technical background.

Two model types

All models in Open Chat Studio fall into one of two categories.

General-purpose models are designed for everyday conversation. They respond quickly and handle a wide range of topics. You shape their answers with a temperature setting.

Reasoning models are designed for tasks that require careful, step-by-step thinking — such as analysis, coding assistance, or complex decision-making. Before producing a final answer, these models work through the problem internally. You control how much thinking they do with an effort setting.

Match your use case

Use this table as a starting point when choosing a model type.

Your chatbot does… Model type to consider
Answer general questions and have conversations General-purpose
Provide friendly support or onboarding General-purpose
Classify or route incoming messages General-purpose
Summarise documents or rewrite content General-purpose
Analyse data, reason through a problem, or write code Reasoning
Handle multi-step workflows requiring careful logic Reasoning
Produce highly accurate factual answers Reasoning (with effort raised)

If you are unsure, start with a general-purpose model and the default settings. You can always switch later — your chatbot configuration is easy to update.

Provider choice

The model you choose determines which AI provider is used. For the full list of supported providers, see LLM Providers.

Your team must have credentials configured for a provider before you can use its models. If the model you want is not listed, check with your team administrator or see the guide for configuring LLM providers.

Different providers may offer the same underlying model type but with different pricing, speed, and availability. If your preferred model is unavailable or too slow, try the equivalent model from another provider.

Next steps