GPT-5.2 Launches: Enterprise AI Tiers Redefined

OpenAI's new strategy for winning the enterprise boils down to a simple, brutal fact: letting an AI think costs real money. GPT-5.2 is that reality made concrete. It arrives not as a single, monolithic brain but as three separate products: Instant, Thinking, and Pro.

This is a company admitting that raw capability is useless if no one can afford the compute to run it. Instant is for the fast, cheap stuff. Thinking is for when you need the model to actually reason, to work through a coding problem or a financial model step by agonizing step.

Pro is for the builders who want to wire the whole thing directly into their systems. The performance claims are sharp. The Thinking model reportedly beats or ties professional humans on 70.9% of tasks like building spreadsheets and presentations.

It also scores 55.6% on the tough SWE-bench Pro software engineering test. The message is clear. They are selling intelligence by the pound.