After nearly two weeks of government-imposed limbo, OpenAI is finally flipping the switch. GPT-5.6 goes public this Thursday, July 9.
The backstory: the Trump administration asked OpenAI to stagger the rollout last month, limiting access to about 20 government-approved partners over national security concerns. OpenAI wasn't thrilled—calling it "not its preferred way" to release new models. But after additional testing by the Commerce Department's Center for AI Standards and Innovation, the green light came through. OpenAI even kept technical experts in D.C. throughout the process to address questions.
Three models, three jobs
Instead of one monolithic release, OpenAI split GPT-5.6 into a family of three:
- Sol — the flagship. Built for heavy lifting: research, software development, multi-step problem-solving. It's priced at $5 per million input tokens and $30 per million output tokens—same as GPT-5.5, but with more capability.
- Terra — the workhorse. GPT-5.5-level performance at roughly half the cost ($2.50 input / $15 output). Everyday writing, analysis, brainstorming.
- Luna — fast and cheap ($1 input / $6 output). Built for high-volume, routine tasks.
What's actually different
The biggest change: GPT-5.6 can think longer before answering. Complex queries get more reasoning time instead of rushed responses. Sol also introduces "Ultra" mode, where multiple AI agents collaborate behind the scenes on complex tasks.
On benchmarks, it's a clear step up. On GeneBench v1 (genomics and computational biology), Sol outperforms GPT-5.5 while using fewer output tokens. In cybersecurity, it's better at finding code vulnerabilities—though OpenAI notes it stops short of carrying out autonomous end-to-end attacks.
All three models are rated "High" risk for cybersecurity and biological/chemical capabilities under OpenAI's Preparedness Framework—the first time even the smaller models hit that mark. None reached "Critical."
The bigger picture
OpenAI isn't alone here. Anthropic went through the same wringer with its Fable and Mythos models. The Trump administration's June AI executive order asks developers to voluntarily submit frontier models for government review before wide release. OpenAI called the process a temporary measure, saying: "We don't believe this kind of government access process should become the long-term default".
For now, though, the models are coming. Thursday can't come soon enough.
Tags:
Technology