OpenAI has replaced its flagship ChatGPT model, GPT-5.6 Sol, with a new one called GPT-6 Astra. It is now live for people on the Pro, Enterprise, and Business Premium plans, and Plus subscribers will get it soon, though only through certain tools rather than the main chat screen.
The most important number in this launch is not a benchmark score. It is the message limit. A Plus subscriber who could send up to 100 messages every five hours with the old model now gets as few as 45 with the new one. That is roughly half the usage for the same monthly fee.
The price for developers building on top of the model went the other direction. Through OpenAI's API, Astra costs 10 dollars per million words of input and 50 dollars per million words of output. The older model charged 2 dollars and 10 dollars for the same amounts. That is roughly five times the cost per unit of text processed.
Why the trade? Astra is a much bigger, more capable system, and bigger systems cost more to run. OpenAI is passing some of that cost onto subscribers and developers rather than absorbing all of it. This is a pattern worth expecting every time a company launches a smarter model: better output, but a smaller ration of it unless you pay more.
What you get for that cost is real, though. Astra can open and edit an actual Excel file, build a dashboard in Power BI, fill out forms in a customer database, and finish computer based tasks close to twice as fast as the model it replaced, according to independent benchmark testing. That is a genuine step toward AI that can do desk work directly instead of just describing what to do.
There is a catch worth sitting with, too. OpenAI itself says this model is harder for its own safety team to keep an eye on, because it works through problems in a new way that is less visible from the outside. It is also the first OpenAI model that triggered the company's toughest internal safety rules, because of how skilled it is at finding and exploiting security flaws. The company's president has even said it might qualify as artificial general intelligence, meaning a system that matches people across a wide range of tasks rather than one narrow skill.
For a business, the lesson is simple. If your team relies on ChatGPT for high volume daily tasks, check whether the new message caps fit your workflow before your Plus plan quietly gets thinner. If you are building automation on the API, price out Astra against your current model before switching, because the jump in cost may outweigh the jump in quality for simple tasks. And if you plan to give this model real autonomy, such as letting it act inside your customer database or file system, keep a person checking its work. Even OpenAI admits it cannot fully see how the model reasons anymore.