OpenAI launches limited preview of GPT-5.6 models Sol, Terra, and Luna following a US government request
Story Overview
OpenAI is giving a small set of trusted partners early access to three GPT-5.6 variants—flagship Sol for complex agentic work and coding, plus Terra and Luna for everyday and high-volume jobs—after previewing the models with the US government, which requested the limited start.
Preview stays behind closed doors for now
The models are available only through the API and Codex to selected organizations; ChatGPT is excluded during this phase, and OpenAI says the government-requested limits are not meant to become the normal pattern.
Sol ships with OpenAI’s strongest safety tools yet
The frontier model adds extra defenses against higher-risk activity, sensitive cyber requests, and repeated misuse while Terra matches prior performance at half the cost and Luna focuses on speed and lowest price.
Positive users express excitement about trying OpenAI's new GPT-5.6 Sol, Terra, and Luna models while negative users object to the limited preview and US-driven restrictions as inaccessible and accuse Dario and Anthropic of cowardice.

how to access it?
Limited preview only: available exclusively to a small group of trusted partners via the OpenAI API and Codex. ¹
Broader access to ChatGPT, Codex, and the API is planned for the coming weeks.

any early data or rumors on if this is better at coding than Fable?
Yes, TerminalBench 2.1 shows GPT-5.6 Sol at 88.8% (Ultra 91.9%) vs. Claude Fable 6 at 84.3%. Early posts highlight Sol's edge in agentic/tool-use and long-running coding/agent tasks, with internal Codex reliance noted.
