OPENAI QUIETLY UNVEILS THREE NEW FRONTIER MODELS BUT ALMOST NOBODY CAN USE THEM YET OpenAI did not make a big announcement. There was no press conference, no splashy blog post with benchmarks and flashy demos. GPT-5.6 just quietly appeared, in three flavors, and almost no one can touch it yet. Sol is the heavy one, aimed at complex coding tasks and security research. Terra is built for high-volume enterprise workloads where speed and cost efficiency matter more than raw capability. Luna is the lightweight option for everyday tasks, designed to be fast and cheap to run. Three distinct models, each carved out for a different slice of the market. About 20 organizations have preview access right now. The US government is apparently among them, which tells you something about the government’s relationship with OpenAI and about who gets to see the bleeding edge before anyone else does. This is not a general release. It is a controlled rollout with monitoring, feedback loops, and the kind of careful staging that OpenAI has learned to do after several high-profile launches that went sideways. General availability is coming in the weeks ahead, the company says. What it means in practice is that OpenAI is already thinking past GPT-5.5, which itself has not been widely available for long. The pace of development is not slowing down. If anything, the cadence of new models is accelerating, and the gap between internal capability and what the public can access keeps widening.