OpenAI announces limited preview of GPT-5.6 models, wider rollout planned

OpenAI has announced a limited preview of its GPT-5.6 AI models for trusted partners, with a broader rollout planned in the coming weeks. The company says the models offer improved reasoning, coding and cybersecurity capabilities amid heightened US regulatory scrutiny.

New Delhi: US-based AI company OpenAI has announced that new GPT‑5.6 models, such as Sol, Terra and Luna, will be rolled out with a limited preview for trusted partners, with a wider release planned in the coming weeks.

These models bring advances in reasoning, coding and cybersecurity, according to the company.

---

OpenAI said it briefed the US government on GPT‑5.6’s capabilities before launch and agreed to the government’s request of a limited launch, while federal authorities develop a broader framework for evaluating advanced models under a recent cybersecurity executive order, according to multiple reports.

The company said it intends to make the models available across ChatGPT, the API and Codex in the coming weeks, and added that it does not want government preview requirements to become a permanent standard for frontier AI releases.

The launch follows increased scrutiny of frontier AI models in the United States.

OpenAI described GPT‑5.6 Sol as its flagship model for the most demanding scientific workloads, Terra as a balanced option for enterprise and developer tasks offering performance comparable to GPT‑5.5 at roughly half the cost.

It launched Luna as the fastest, most affordable model for cost‑sensitive applications without sacrificing core capabilities.

The company said Sol will be available on Cerebras hardware in July for select customers, delivering inference speeds up to 750 tokens per second.

Sol can use Max Reasoning Effort, a capability upgrade, to spend significantly more time reasoning through difficult problems before responding. Its Ultra Mode, deploys multiple specialised sub-agents that work together on complex workflows, improving performance on coding, research and multi-step tasks.

Meanwhile, US-based AI company Anthropic won US approval to restore limited access to its powerful Mythos 5 artificial intelligence model to “certain trusted partners” after resolving the government’s national security concerns.

The clearance follows a government order two weeks earlier that abruptly barred Anthropic from giving foreign nationals access to Mythos 5 and a related model, Fable 5, over fears that security guardrails could be circumvented.