AI providers & models supported by GPT for Work
GPT for Work supports models from Anthropic, DeepSeek, Google, Mistral, OpenAI, OpenRouter, Perplexity, and xAI. GPT for Work also supports open-source models through Ollama and other local servers with an OpenAI-compatible API. The tables below show which models you can use with and without your own API key, in which GPT for Work add-ons, and at what price.
Models you can use without your own API key
Models you can use with your own API key
You pay the API cost directly to the AI provider.
For past changes in model pricing, see Price evolution.
Ensure the AI models you use comply with your local laws and suit your needs.
DeepSeek should not be used in regions where its use is prohibited. When its use is allowed, avoid sharing personal, confidential, or sensitive information.
Notes
Web search models
Some models can search the web for up-to-date information to use as context when generating responses. Moreover, some models allow you to choose how much context they gather. This is measured as context size, which ranges from low to high. Larger context sizes allow models to retrieve more information from each web source, which can produce richer and more nuanced responses.
In addition to the regular token cost, web search models incur a separate search cost when you do not use your own API key. The search cost is billed per 1,000 searches. For models that allow you to choose the amount of context, larger context sizes consume more of your balance.
The following table lists the context size costs for each supported web search model when not using your own API key, billed per 1,000 searches.
Provider | Model | Low | Medium | High |
---|---|---|---|---|
gemini-2.5-flash | $35.00 (fixed context size) | |||
Perplexity | sonar | $5.00 | $8.00 | $12.00 |
Perplexity | sonar-pro | $6.00 | $10.00 | $14.00 |
Reasoning models
Reasoning models split output tokens into completion tokens and reasoning tokens. Completion tokens are the tokens that make up the model's answer, while reasoning tokens are additional tokens generated during the model's reasoning process. You are billed for both types of tokens.
Local servers
The set of supported models depends on the locally running LLM server you are using. You can use any local server that provides an OpenAI-compatible API, such as LM Studio, KoboldAI, LocalAI, or Open WebUI.
The tables on this page were created with Awesome Table Apps.