OpenAI ships GPT-5.6 behind government review
On June 26 OpenAI opened a preview of GPT-5.6: Sol, Terra and Luna, gated to about 20 organizations under a federal pre-release review.
Mariano De Vitto · June 2026
On June 26 OpenAI began a limited preview of GPT-5.6, a three-model series named Sol, Terra and Luna. The release came gated. During the preview it is available only to about 20 trusted organizations, through the OpenAI API and the Codex coding tool, under a federal pre-release review framework created by a June 2 executive order. Individual ChatGPT users cannot use it yet, and general availability is expected in the coming weeks.
The series splits into three purpose-built tiers. Sol is the flagship, OpenAI's most capable model yet, aimed at high-stakes and cybersecurity work such as vulnerability research and long-horizon reasoning, and it runs in two heavier modes, max for deeper single-model reasoning and ultra for parallel sub-agents. Terra is the balanced model for everyday work, with performance competitive with GPT-5.5 at about half the cost. Luna is the budget tier, fast and affordable, built for high-volume routine tasks. Pricing per million tokens runs 5 and 30 for Sol, 2.50 and 15 for Terra, and 1 and 6 for Luna.
Two changes matter for builders. Sol's context window is reported at about 1.5 million tokens, roughly 43 percent larger than the GPT-5.5 generation, though OpenAI's preview focused on safety rather than the spec sheet, so treat that figure as reported until general availability. Prompt caching is now more predictable, with explicit cache breakpoints and a 30-minute minimum cache life, which lowers cost on repeated context.
What it means for marketing
The bigger story is access. GPT-5.5 shipped to everyone on day one. GPT-5.6 ships to a short list, reviewed by the government before release, the first time a major model has launched this way. That is a precedent about where regulation is heading, and a reminder that the tools you build campaigns on can have their availability changed by policy, not only by the vendor.
Plan for capability that arrives in stages. Watch the access calendar as closely as the benchmarks, keep a second provider live for workflows you cannot pause, and pick the tier by the job: Sol for security-sensitive or complex reasoning, Terra as the default for most content and production, Luna for bulk routine volume. Re-test when the gate widens to general availability later this summer.
The signal: The strongest model now ships gated, reviewed and split into three priced tiers. Match the tier to the task, keep a backup provider, and treat access as a variable, not a constant.
The Signal Brief · Mariano De Vitto — Head of Marketing, Barcelona