Three companies, one day. GitHub Copilot, Figma Make and Framer (a site-building tool) announced support for GPT-5.6 on 9 July - a model in three variants: Sol, Terra, Luna. It reached the coder, the designer and the site builder within almost the same hour.
Fact
The facts: Sol is the top tier - for complex reasoning and agentic work over large codebases. Terra is the balanced option for everyday work. Luna is the light, cheapest one, for small and fast tasks. GitHub Copilot rolls them out across 10 environments, including VS Code, JetBrains, Xcode, github.com and GitHub Mobile. By Figma's own account, their early tests show GPT-5.6 raising both the quality and the speed of the first result in Make. Framer claims Sol is the first model to score 100% on their hardest internal benchmark, that Terra matches GPT-5.5 results at 0.6x credits, and that Luna is twice as fast at 0.4x credits. All three announcements are from 9 July 2026.
Here's the substance. The news here isn't the new model. The news is the speed it rolls out at - into the coder's, the designer's and the site builder's environment within a single day. And the structure has changed: you no longer buy a model, you pick a tier - powerful, mid, cheap. The question shifts from which model to how much this task costs.
A hundred percent on your own benchmark is still your own result - for now.
What occupies me less is the benchmark, more the address. Figma Make and Framer are where form gets born - that's where you sketch, tear up and start over. A model dropped straight in there sits at the sketching table, instead of waiting in the menu as an option. Whether the promised higher quality of the first result survives a week of real use, remains to be seen. The day of the announcement proves nothing.
The ladder of three models admits something simple - not every task deserves the most expensive tool. The light one for quick work, the heavy one for the hard stuff. That's mature thinking about real use. I'll judge by what's still in the workflow a month from now.
The visual is generated code-art, no third-party images.