On June 27 local time, OpenAI officially released preview versions of its GPT-5.6 series models. The new lineup includes the flagship Sol model, mid-tier Terra, and cost-effective lightweight Luna, targeting high-performance reasoning, high-throughput tasks, and everyday AI applications respectively. Just one day prior, media reports indicated that at the request of the Trump administration, OpenAI would adjust its release schedule for these new models and initiate limited testing first. OpenAI stated that the GPT-5.6 series significantly enhances capabilities in specialized domains such as code generation, cybersecurity, and biology, while also improving sustained reasoning and execution abilities during long-cycle agent tasks. Among them, the flagship Sol is priced at $5 per million input tokens and $30 per million output tokens; Terra costs approximately half of Sol's price, while Luna further reduces expenses to serve broader developer and enterprise users. Beyond base versions, OpenAI introduced two additional operating modes for Sol: "Max" mode designed for complex reasoning tasks and "Ultra" mode capable of coordinating multiple sub-agents, thereby expanding its ability to handle intricate workflows. Security remains central to this launch. According to OpenAI, GPT-5.6 has been specifically trained to reject malicious cyberattack requests, including attempts involving jailbreaking the model or disguising true intentions. The company noted that Sol excels at helping developers identify and patch security vulnerabilities rather than executing full-scale attack procedures, thus failing to meet the critical risk level defined by its internal Preparedness Framework for cybersecurity threats. To validate model safety, OpenAI revealed it invested approximately 700,000 A100e GPU hours in automated red team testing and invited third-party security researchers to participate in ongoing evaluations, with tests continuing over the next two weeks. Additionally, due to close U.S. government oversight during the current preview phase, some legitimate dual-use scenario requests related to cybersecurity may be mistakenly blocked under existing security policies—a key aspect being verified through this preview test. OpenAI expects the GPT-5.6 series will become widely accessible to more users within the coming weeks. Emphasizing its commitment to broad accessibility, the company highlighted that this limited preview implemented in collaboration with the U.S. government represents an exceptional arrangement. It expressed hope that future model releases can adopt standardized, replicable approval processes instead of relying on prolonged case-by-case government reviews.