OpenAI opens limited access to GPT-5.6 Sol, Terra, and Luna: analysis of the new model family
OpenAI has taken an important step in the development of its technologies by opening limited access to the new family of models GPT-5.6 Sol, Terra, and Luna. Access is provided via API and Codex to a narrow circle of trusted partners. This preview, it is noted, was initiated at the request of US authorities, highlighting the growing role of regulatory oversight in the field of AI.
It is planned that the models will become publicly available in the coming weeks. However, at the initial stage, OpenAI pre-agreed the plans and capabilities of the models with the US government, and also provided information about the partners. The company emphasizes that it does not consider this procedure as a long-term norm, indicating a search for a balance between innovation and safety.
Lineup and Performance
Sol is positioned as the flagship model, Terra as a "workhorse" for everyday tasks, and Luna as a fast and economical option. According to OpenAI estimates, Terra demonstrates performance comparable to GPT-5.5, but at half the cost. Luna, in turn, is the most affordable model in the lineup.
GPT-5.6 Sol is the company's most powerful model to date. It is equipped with a max reasoning mode, which provides more time for deep processing of complex tasks, and an ultra mode, which uses sub-agents to accelerate complex work. This allows Sol to solve tasks requiring both deep analysis and high processing speed.
Tests and Safety
According to OpenAI, Sol set a new record in the Terminal-Bench 2.1 benchmark for command-line tasks. On GeneBench v1, the model showed results higher than GPT-5.5 with fewer tokens. In ExploitBench tests, Sol proved competitive with Mythos Preview, using approximately one-third of the output tokens. On ExploitGym tests, all three models improved results in cyber tasks as reasoning depth increased.
Safety is a key aspect. OpenAI states that GPT-5.6 Sol, Terra, and Luna received the strongest safety stack in the lineup. Sol does not exceed the Cyber Critical threshold under the Preparedness Framework. In tests with Chromium and Firefox, the model found bugs and vulnerability exploitation primitives but could not autonomously create a fully functional working exploit. More than 700,000 GPU-hours in A100 equivalent were used for automated red teaming to verify protection. During the preview stage, multi-layered measures are in place, including model-level restrictions, real-time checks, and monitoring.
Pricing and Plans
New naming scheme: the number denotes the generation, while Sol, Terra, and Luna denote capability levels. The price per 1 million tokens is: for Sol — $5 input and $30 output, Terra — $2.50 and $15, Luna — $1 and $6.
In July, OpenAI plans to launch GPT-5.6 Sol on the Cerebras platform with speeds of up to 750 tokens per second. Access at this stage will also be limited to select clients while the company expands capacity.
My analysis: The launch of GPT-5.6 Sol, Terra, and Luna is not just another update, but a strategic move by OpenAI to segment the AI market. Dividing into flagship, workhorse, and budget models allows the company to cover different user segments, from large corporations to small developers. However, the limited access at the request of US authorities signals that even industry leaders must consider regulatory risks. In the coming months, we will likely see how these models impact the competitive landscape, especially in the areas of cybersecurity and task automation.