AI models

GPT Workspace picks the model for each job and charges credits at the model’s actual cost, see What is a credit?. You choose a tier in Sheets formulas and an image model in =GPT_IMAGE; the chat agents select their models themselves.

Tier	Model	Plans
Standard	GLM 5.3 Flash	All plans
Advanced	GLM 5.3 Flash, where newer and heavier models are added first	Standard and Pro

Both tiers are kept up to date with current models. The tier is set per formula with the model parameter, =GPT("…", A1, , "advanced"), or for all formulas in Formula settings.

=GPT_CLASSIFY uses a dedicated decision model (TypeSafe Jev) that is about ten times cheaper than a text formula.

The chat runs a main agent and one specialist per Google app. They use the current frontier models (GPT, Claude, Gemini and GLM families) chosen per task: a fast model for routing and short answers, a stronger one for writing and planning, a multimodal one for files and images. Advanced models in chat are available on every paid plan.

=GPT_IMAGE and the chat’s image generation let you pick the model:

model value	Model	Relative cost	Plans
gemini (default)	Gemini 3.1 Flash Lite Image (nano banana)	lowest	All plans
gpt	GPT Image 1 mini	low	All plans
openai/gpt-image-2	GPT Image 2	high	Paid plans
google/gemini-3-pro-image	Gemini 3 Pro Image	highest	Paid plans

Legacy names (dall-e, dall-e-3, gpt-image-1, gpt-5-image) keep working and map to the closest current model.

=GPT_WEB and the chat’s web search use Perplexity Sonar Pro Search, which has live web access. =GPT_WEB is available on paid plans; because the provider bills per search whatever the length of the answer, each call carries a minimum charge.

=GPT_FILE reads PDFs, images, audio and video with a multimodal Gemini model, and indexes other documents for search, see GPT_FILE.