OpenAI has released a preview version of the GPT-5.6 model, currently available only to trusted Codex and API partners

BlockBeats News, June 27th. According to official sources, OpenAI has released a preview version of the GPT-5.6 series models, including three different specifications: Sol, Terra, and Luna. Due to the U.S. government's restrictions on advanced AI model security review framework, the models are initially only open to a few trusted partners in a limited preview and are planning to be released to the public in the coming weeks.

The flagship model Sol introduces the Ultra mode, which combines maximum inference strength with sub-agent collaboration. In the Terminal-Bench 2.1 command line workflow test, Sol scored 88.8%, and the score increased to 91.9% in Ultra mode, surpassing GPT-5.5 at 83.4% and Claude Fable 5 at 88.0%. The mid-range model Terra approaches GPT-5.5 in performance and is priced at half the cost, while the lightweight model Luna is designed for everyday automation tasks. Sol is priced at $5 per million input tokens, $30 per output, and supports using prompt caching to reduce the cost of secondary calls.

In terms of security, a security assessment confirmed that Sol has not crossed the Preparedness Framework's network security critical threshold. OpenAI has invested over 700,000 A100-equivalent GPU hours in automated red team exercises, equipping the entire model lineup with a defense stack that includes a denial mechanism, real-time abuse classifier, and account-level audit. While the current limited release follows the U.S. government's security framework, OpenAI emphasizes that they do not want a government-led access mechanism to become a long-term default mode, as it would restrict defense personnel from accessing cutting-edge tools.

---------------------------------
Click the original article link below to join the Epsilon Beatings Lark AI News Channel and receive 24/7 monitoring of global AI hotspots and news.