What happened
On Thursday, OpenAI announced GPT-5.6, a model family with three variants. Sol is the workhorse, Terra is the intermediate option, and Luna is the budget-friendly pick. The company is promising strong results across enterprise work, coding and scientific research.
The headline number comes from the boss. Sam Altman told CNBC that Sol is 54% more token efficient on AI coding tasks. He also says the new models are far more efficient and cost-effective than their predecessors.
Then there is security. OpenAI describes 5.6 as its strongest cybersecurity model yet, reaching frontier-level performance while using significantly fewer tokens.
The model's cyber abilities have already drawn plenty of noise. The Trump administration previously tried to restrict the rollout, reportedly out of concern about misuse. OpenAI says GPT-5.6 supports defensive work: threat modeling, code review and patching, and blue teaming, which means simulating attacks on your own systems.
Why it matters
A three-tier lineup is a pricing strategy wearing a space costume. Developers and companies get a clear ladder: pay more for Sol, pay less for Luna, and pick the rung that fits the job.
The efficiency pitch matters just as much. If Sol really needs fewer tokens to finish coding work, that is the kind of gain teams feel on the invoice, not only on a leaderboard. Keep in mind the 54% figure is a claim from OpenAI's CEO, not an independent test.
The cyber angle is the spicy part. A model that can find and patch holes can also, in theory, help someone dig them. That tension is why regulators noticed before the launch event did.
The cyber pitch is defensive, but the model that defends is also the one governments worried about.
The naming tells you how OpenAI now thinks about models: not one assistant, but a family, with a heavy model for hard thinking and lighter ones for speed and price. For users, the practical question is which one answers by default and which ones sit behind a paywall. For developers, it is whether the cheaper tiers are good enough to switch the bill.
What to do with it
- Check the tier.Match your workload to Sol, Terra or Luna before defaulting to the biggest one.
- Run your own coding test.Measure token use on your real tasks, because 54% is OpenAI's number, not yours.
- Loop in your security team.If defensive work like code review and patching is on your list, see where the new model fits.
- Watch the pricing page.Efficiency claims only count once they show up in your bill.
Ciao, sapiens. The sun, the planet and the moon are all working nights, so you can sleep.


