OpenAI opens a limited preview of GPT-5.6 Sol, Terra and Luna

OpenAI has begun a limited preview of the GPT-5.6 series: Sol, the flagship model; Terra, a balanced model for everyday work that the company says matches GPT-5.5’s performance while being twice as cheap; and Luna, a fast and affordable tier. The company says the family should become generally available in the coming weeks.

Government coordination and phased access

OpenAI says it previewed its plans and the models’ capabilities with the U.S. government before launch and, at the government’s request, is starting with a small group of trusted partners whose participation has been shared with officials. The company states that it does not believe this kind of government access process should become the long-term default, because it keeps the best tools from users, developers, enterprises, cyber defenders and global partners. It frames the step as the fastest path to broader availability while it works with the Administration on a cyber executive-order framework and a repeatable process for future model releases.

New reasoning modes and benchmark results

GPT-5.6 introduces a new “max” reasoning effort that gives Sol more time to think, and an “ultra” mode that goes beyond a single agent by using subagents to accelerate complex work. In coding, OpenAI says Sol sets a new state of the art on Terminal-Bench 2.1, which tests command-line workflows requiring planning, iteration and tool coordination. On GeneBench v1, a long-horizon genomics and quantitative-biology benchmark, it outperforms GPT-5.5 while using fewer tokens. In cybersecurity, Sol is competitive with Mythos Preview on ExploitBench using roughly one-third of the output tokens, and the whole family shows strong gains on ExploitGym, a benchmark created by UC Berkeley researchers with OpenAI and other frontier labs.

Safeguards, pricing and a Cerebras launch

OpenAI says Sol does not cross the Cyber Critical threshold under its Preparedness Framework: in Chromium and Firefox evaluations it identified bugs and exploitation primitives but did not autonomously produce a functional full-chain exploit. The preview relies on layered safeguards — model-level refusals, real-time cyber and biology classifiers that can pause generation for review by a larger reasoning model, account-level signals and differentiated access — plus more than 700,000 A100-equivalent GPU hours of automated red teaming aimed at universal jailbreaks. Pricing per million tokens is $5 input and $30 output for Sol, $2.50 and $15 for Terra, and $1 and $6 for Luna, with explicit prompt-cache breakpoints and a 30-minute minimum cache life. Sol will also launch on Cerebras at up to 750 tokens per second in July, initially for select customers.