OpenAI unveiled a limited preview of its GPT‑5.6 series, bundling three models: the flagship GPT‑5.6 Sol, the general‑purpose Terra, and the budget‑oriented Luna. Sol adds agentic reasoning, new “max” effort and an “ultra” mode that deploys sub‑agents for complex tasks. OpenAI aims to widen access in weeks after further safety checks.

Terra matches GPT‑5.5 performance while costing 2x cheaper, positioning it as the workhorse for enterprises. Luna promises “strong capability” at OpenAI’s lowest price point, targeting high‑volume use cases. OpenAI claims Sol carries its most robust safety stack yet, with protections against high‑risk cyber requests and hardened defenses tested against real‑world attacks, favoring defensive over offensive applications, making it attractive for cost‑sensitive workloads.

The Trump administration invoked a June 2 executive order to benchmark new AI models, prompting OpenAI to restrict GPT‑5.6 to trusted partners for now. OpenAI pushed back, arguing that permanent government vetting would choke innovation for developers and cyber defenders. The company says a wider rollout to ChatGPT, Codex and the API will follow once testing concludes, ensuring compliance with emerging regulations.