OpenAI has taken a significant step in the development of its language models by providing limited access to the new GPT-5.6 family via API and Codex. The lineup, consisting of three models — Sol, Terra, and Luna — is currently available only to a narrow circle of trusted partners. This launch is taking place in a preview format in coordination with U.S. authorities, highlighting the growing role of regulatory aspects in the AI industry.
OpenAI plans to make GPT-5.6 publicly available in the coming weeks, but at the initial stage, the company has taken unprecedented measures: all partner data has been handed over to the government, and the launch is restricted. The company itself emphasizes that this procedure is not a long-term norm, but it demonstrates how seriously the risks associated with powerful AI systems are being taken.
Three Performance Levels
In the new lineup, Sol is positioned as the flagship model, Terra as the "workhorse" for everyday tasks, and Luna as the fastest and most cost-effective option. According to OpenAI, Terra delivers performance comparable to GPT-5.5 but at half the cost. Luna becomes the most affordable model in the family, opening up new opportunities for developers with limited budgets.
Sol, according to the company, is the strongest OpenAI model to date. It features two new reasoning modes: max, which gives the model more time for deep processing of complex tasks, and ultra, which uses sub-agents to accelerate the execution of complex operations. This is a fundamentally new approach to AI architecture that could reshape the market.
Performance Records and Safety
Test results are impressive. Sol set a new record in the Terminal-Bench 2.1 benchmark for command-line tasks, and on GeneBench v1, it outperformed GPT-5.5 with fewer tokens. On ExploitBench, the model proved competitive with Mythos Preview while using only about a third of the output tokens. This indicates a significant increase in efficiency.
Special attention has been paid to safety. OpenAI states that GPT-5.6 Sol, Terra, and Luna have received the most powerful protection stack in the lineup. The model does not cross the Cyber Critical threshold within the Preparedness Framework, and in tests with Chromium and Firefox, it found bugs but could not autonomously create a working exploit. To verify protection, the company used over 700,000 GPU-hours (in A100 equivalent) for automated red-teaming.
Pricing is also noteworthy: for 1 million tokens, Sol is priced at $5 for input and $30 for output, Terra at $2.50 and $15, and Luna at $1 and $6. OpenAI has already announced plans for July — the launch of GPT-5.6 Sol on the Cerebras platform with speeds of up to 750 tokens per second, though access at this stage will also be limited.
My analysis: The launch of GPT-5.6 marks a shift toward finer segmentation of the AI model market. The division into flagship, workhorse, and budget versions is a direct response to business demands seeking a balance between cost and performance. However, the key point is regulatory pressure. The fact that OpenAI agreed to a limited launch at the request of U.S. authorities could set a precedent for the entire industry. In the coming months, we will likely see other major players forced to follow similar protocols, which will slow adoption rates but increase trust in the technology.