OpenAI Rolls Out GPT-5.6 Sol, Terra and Luna After White House Pressure

OpenAI began the public rollout of its GPT-5.6 model family, Sol, Terra and Luna, on July 9, following two weeks of limited testing with trusted partners. The company acknowledged that the rollout's pace was not entirely its own call, but the result of prior arrangements with the US administration.

Three models, one version number

Sol is the flagship of the family, built for harder tasks such as complex reasoning, coding, scientific work and long, autonomous agentic tasks. OpenAI describes it as a major leap over GPT-5.5. Terra is meant to match the previous generation's performance at half the cost, while Luna is the fastest and cheapest option, designed for lighter workloads.

The names reference the Sun, Earth and Moon, which quickly sparked jokes on social media since 'Sol' overlaps with the ticker for the Solana cryptocurrency. Beyond the marketing mix-up, the model brings real technical additions: Sol gets a new, top-tier reasoning effort level called max, plus an ultra mode in which the model delegates part of the work to subagents to speed up complex tasks.

In Terminal-Bench 2.1 testing, which measures command-line work, the Sol Ultra variant scored 91.9 percent, a result OpenAI points to as evidence of its edge in agentic tasks involving coding and system administration.

Why the rollout is staggered

The most unusual part of the launch, though, isn't the models themselves but how they're being released. Sam Altman told employees that the gradual, staggered rollout of GPT-5.6 came at the government's request. As part of its ongoing engagement with the US administration, OpenAI had reportedly demonstrated the new models' capabilities in advance, and at officials' request began with limited access for a narrow group of trusted partners whose involvement was disclosed to the government.

The White House stressed that this wasn't a formal approval process or top-down control over the launch, OpenAI did not need to obtain any official clearance to release the models. Still, the mere fact that an AI lab is coordinating its release schedule with the government shows how much the regulatory environment around the most powerful models has shifted over the past year.

Sol is our new flagship model and a step change from GPT-5.5. Terra delivers performance competitive with GPT-5.5 at half the cost. Luna is our most cost-efficient model, offering solid capability at the lowest price. - OpenAI, GPT-5.6 launch announcement

High risk ratings

The system card published alongside the launch rates all three models, Sol, Terra and Luna, as High in two risk categories: biological/chemical threats and cybersecurity. That is currently the highest category OpenAI applies before a threshold that would require pausing deployment. In cybersecurity testing, Sol and Terra were able to find vulnerabilities and exploit fragments, but could not carry out a full attack on a well-secured target on their own.

For Polish companies and developers, availability matters more than capability right now. During the testing phase, Sol, Terra and Luna are available only through the API and Codex, and only to a select group of partners, they are not yet in ChatGPT. OpenAI says wider availability in the API, Codex and ChatGPT is coming in the following weeks, without giving a specific date.

What it means for the market

The GPT-5.6 launch coincided with moves from other leading labs: Anthropic's current default model is Claude Sonnet 5, and Google DeepMind pushed back the launch of Gemini 3.5 Pro to July 17 while rebuilding the model's architecture from the ground up. For companies using the API, this means several new flagship model generations will hit the market within weeks of each other, which usually triggers another round of price cuts and contract renegotiations with providers.

The growing practice of coordinating release schedules with the government, now visible with GPT-5.6, could set a precedent for future rollouts of the most powerful models, regardless of whether the United States establishes a formal legal basis for such oversight or it remains an informal industry practice.

Sources: OpenAI Help Center - A preview of GPT-5.6 Sol, Terra, and Luna (help.openai.com), Neowin - OpenAI to release GPT-5.6 Sol, Terra and Luna on July 9 (neowin.net), Yahoo Tech - GPT-5.6 Is Here (tech.yahoo.com)