OpenAI on Friday unveiled the GPT-5.6 family of AI models, launching a limited preview of its new models codenamed Sol, Terra, and Luna before a broader planned release expected in the coming weeks.
The announcement comes one day after reports that President Donald Trump’s administration asked OpenAI to limit GPT-5.6's initial release while officials evaluate the model under a developing federal framework for frontier AI systems—reports that proved accurate.
Calling it the flagship model, OpenAI said GPT-5.6 Sol improves performance across coding, biology, and cybersecurity. The company said Terra delivers performance comparable to GPT-5.5 at a lower cost, while Luna is designed for high-volume, low-cost workloads. The release also introduces "max" and "ultra" reasoning modes, giving Sol more time to solve complex problems or coordinate multiple subagents for demanding tasks.
“We're beginning a limited preview of the GPT‑5.6 series: Sol, our flagship model; Terra, a balanced model for everyday work; and Luna, a fast and affordable model,” OpenAI wrote. “Terra has competitive performance to GPT‑5.5 while being 2x cheaper, and Luna brings strong capability at our lowest cost.”
In the announcement, OpenAI confirmed it shared the models with the U.S. government before launch, and is beginning with a limited preview at the administration's request while the two sides develop a process for future frontier AI releases.
“As part of our ongoing engagement with the U.S. government, we previewed our plans and the models’ capabilities ahead of today’s launch,” the company wrote. “At their request, we are starting with a limited preview for a small group of trusted partners whose participation has been shared with the government, before releasing more broadly.”
The move follows the administration's earlier order that Anthropic limit access to its Fable 5 and Mythos 5 models, making GPT-5.6 the second frontier AI system this month whose rollout has been impacted by the White House.
In testing, OpenAI claimed Sol achieved the highest score on TerminalBench, a benchmark for command-line software engineering tasks, and outperformed GPT-5.5, Claude Mythos 5, and Fable 5.
“GPT‑5.6 Sol also shows broad improvements in biology workflows,” OpenAI wrote. “On GeneBench v1, which evaluates long-horizon genomics and quantitative-biology analyses, it achieves stronger results than GPT‑5.5 while using fewer tokens.”
In terms of cybersecurity, OpenAI said GPT-5.6 pairs stronger cybersecurity capabilities with expanded safeguards designed to support defensive security research, while limiting offensive misuse. OpenAI added that the model remains below its Cyber Critical threshold because, although it can identify vulnerabilities and exploit components, it could not autonomously produce a complete exploit chain during testing.
“GPT‑5.6 is trained to refuse prohibited cyber assistance, including when users attempt to disguise their intent or jailbreak the model,” the company said. "These model-level safeguards establish the first boundary around what the model should and should not help with.”
According to OpenAI, during the preview, GPT-5.6 Sol, Terra, and Luna will be available through the API and Codex to a select group of partners before expanding publicly to ChatGPT and other users. OpenAI also introduced a new naming system for the model family and said GPT-5.6 Sol will launch on Cerebras in July, offering inference speeds of up to 750 tokens per second.
Despite the ordered limited rollout, OpenAI said it still intends to make the models broadly available to the public.
“We believe in broad access, and we plan to make GPT‑5.6 Sol, Terra, and Luna generally available in the coming weeks,” they said.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
