GPT-4o

Why different input types matter

OpenAI described GPT-4o as a model trained across text, vision and audio. The launch also involved staged access, so the model’s described abilities and the features available in a particular product were not identical.

For a user, the value of an input type is the work it removes. A photograph can save someone from manually typing a table. Spoken interaction can make a conversation easier while their hands are occupied. These benefits need their own checks: transcription errors, image quality and interruptions can matter more than a broad reasoning score in the workflow being evaluated.

GPT-4o launch announcement

From an image to something usable

Use a fictional café menu as an example. Ask a model to extract item names and prices into a table. Check every cell against the image before asking it to calculate a total. If a price is unreadable, the useful behavior is to mark it as uncertain, not fill the gap with a plausible amount.

A second task might ask for a concise description of the menu’s layout for someone who cannot see it clearly. That requires different scoring from extracting prices. A description can omit an irrelevant decorative flourish and still be useful, while a missing decimal point may ruin the table. Both examples illustrate potential workflows rather than measured performance on this site.

A more useful comparison than one overall winner

For this release, compare input support, response quality, availability and the time needed to finish a task separately. A fast first answer may require several corrections. An apparently slower answer may need less editing. If you measure speed, include the human work required to reach an acceptable result.

Keep historical prices and current prices separate too. This page explains a 2024 announcement; it should not be used as a live purchasing guide. The current model catalogue can help locate later releases, but a buying decision still needs today’s product access and pricing checked at the source. The same distinction applies when a model name remains familiar while the product around it changes.