OpenAI released its GPT-5.6 family on 26 June. Three models: Sol, the flagship, aimed at hard agentic coding and scoring 88.8% on the Terminal-Bench 2.1 benchmark. Terra, a balanced everyday option. And Luna, the one most South African businesses should care about: fast, and cheap at roughly one dollar per million input tokens and six dollars per million output tokens.
Then came the catch. The US government asked OpenAI to limit the launch to around 20 approved organisations while federal officials review the models for cybersecurity risk. Access is granted one customer at a time. A wider release is planned, but there is no firm date.
Why the Restriction Is a Big Deal
This is the first time the US government has asked an American AI company to hold back a model before release rather than after a problem appeared. It happened in the same month the government pulled Anthropic's Fable 5. Read together, the message is clear: frontier model launches now pass through a government gate, and that gate is in Washington.
Why Luna Matters Here
Most business AI work is not the hardest reasoning task. It is volume: answering the same customer questions, sorting enquiries, drafting first versions, tagging data. For that work, the cost per token decides whether a feature is worth building. Luna's pricing makes things that were borderline last year clearly affordable. A chatbot that handles a few thousand conversations a month, or an automation that reads every inbound email, gets cheaper to run.
You cannot build on Luna today from South Africa. But the direction is set. When access opens, the cost of adding AI to a small business drops again.
What to Do Now
Do not sit idle waiting for it. Build what you need on the models available now, the current GPT and Claude tiers, and keep the model choice in one place in your setup so you can switch to Luna the day it opens. The teams that already have a working AI feature will swap in a cheaper model in an afternoon. The teams starting from zero will still be planning.