Manage models for GPT for Excel

Models are the engines that power GPT for Excel. GPT for Excel supports models from OpenAI, Perplexity, Google, Anthropic, Parallel, DeepSeek, Mistral, and xAI. GPT for Excel also supports Azure models and open-source models through any OpenAI-compatible API endpoint.

info

As a space owner or admin, you can control which models are available to users in your space through the AI configuration section in the GPT for Work dashboard.

Which model is the right one depends on your specific needs in terms of response quality, speed, privacy, and cost. The following table summarizes the kinds of models available in GPT for Excel, from zero setup to more advanced options:

Use	Setup	When
Models without an API key	None. Use the default model or select a different model: Agent Bulk tools GPT functions	You want to get started quickly without setting up API keys. ✅ Available models
Models with an API key 🔑	AI provider models Set an API key for an AI provider.	You want to benefit from the terms of service you have with an AI provider, such as cost, privacy, and rate limits. You want to use a model that is not available without an API key. You want more control. ✅ Available models
Models from cloud endpoints	Azure OpenAI models Set up an Azure OpenAI endpoint.	Your company is already using Azure with a consumption commitment. Using Azure OpenAI counts towards that commitment. You want to use models that are available through Azure OpenAI. ✅ Available models
Models from cloud endpoints	OpenAI-compatible models Set up an OpenAI-compatible platform endpoint.	You want to use models available on a cloud-based LLM platform such as Anyscale or Fireworks AI. You need a fast and powerful inference infrastructure that scales to your performance needs.
Local models	OpenAI-compatible models Set up an OpenAI-compatible server endpoint.	You want to use models available on a local LLM server such as LM Studio or Ollama. You want to run models on a local machine. You do not want to send or receive any prompt or response data over the internet.

What's next