Overview
- OpenAI previewed the GPT-5.6 family on June 26 and introduced three tiers called Sol, Terra, and Luna, with Sol described as the most capable for complex tasks and cybersecurity work.
- Access is restricted to a small group of vetted partners at the U.S. government's request, with reports saying roughly 20 organizations have been granted API and Codex preview access.
- OpenAI says GPT-5.6 uses a multi-layer safety stack — trained refusals, real-time classifiers that pause risky outputs, account-level reviews, and automated plus human red‑teaming that ran hundreds of thousands of GPU hours — to limit dual‑use misuse.
- The company published per‑token pricing for each tier and said it plans broader availability to ChatGPT users, Codex customers, and the wider API in the coming weeks once the government review process proceeds.
- The model names triggered short‑lived crypto trades (notably a spike in LUNA2 futures), and the rollout is prompting developers to build product routing, fallback plans, and clearer audit trails as government vetting becomes a new release constraint.