OpenAI GPT-5.6 launch: Sol, Terra, Luna pricing details

OpenAI unveiled GPT-5.6 on Jun 26, 2026, releasing a limited preview of a three-model suite called Sol, Terra, and Luna. The company positioned Sol as the flagship, Terra as a mid-tier option for "high-volume work," and Luna as a fast, affordable everyday model.

What is included in the GPT-5.6 suite?

Sol, Terra, and Luna make up GPT-5.6, with Sol described as the flagship and Sol specially tuned for coding, cybersecurity, and biology. Sol also includes two additional modes: a "max" mode for deeper reasoning and an "ultra" mode for leveraging sub-agents. The release notes say GPT-5.6 is "especially skilled at coding, cybersecurity, and biology, as well as staying focused during long-horizon agentic AI tasks."

Terra is positioned as a medium-tier model for high-volume work, and Luna is framed as a "fast and affordable" everyday model. The announcement ties Sol’s advanced capabilities to developer tooling for complex, multi-step workflows and to sub-agent orchestration via the new ultra mode, a feature that echoes references to OpenClaw and OpenClaw creator Peter Steinberger’s work at OpenAI.

How is OpenAI pricing and positioning GPT-5.6?

OpenAI published per-million-token pricing for the suite: GPT-5.6 Sol is priced at $5 input and $30 output per million tokens. The company says Terra is half the cost of Sol, while Luna is less than half the cost of Terra. OpenAI contrasted Sol’s pricing with Anthropic’s Claude Fable 5, noting Claude Fable 5 is priced at $10 input and $50 output per million tokens.

The pricing framing places Sol as a high-performance but lower-cost flagship compared with a direct competitor, Terra as a cost-reduced option for volume use, and Luna as the cheapest tier for everyday tasks. OpenAI also signaled a broader availability timeline, saying the model suite should be generally available in the coming weeks while the company runs the preview.

How is OpenAI handling safety and the regulatory review?

OpenAI dedicated most of the announcement to safety and misuse risk, saying Sol has the company’s "most robust safety stack to date" and that it is trained to refuse prohibited cyber assistance, including when users attempt to disguise their intent or jailbreak the model. The company claimed Sol is "better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks," and that Sol does not cross the cyber-critical threshold under OpenAI’s preparedness framework.

OpenAI said it devoted approximately 700,000 A100e GPU hours to automated red-teaming and that it worked with third-party testers, who will continue testing the model for two more weeks. The preview period is being monitored by the Trump administration, and an earlier report said the administration will approve customers on a case-by-case basis during the preview. OpenAI wrote that it cooperated with the US government ahead of launch and added, "We don’t believe this kind of government access process should become the long-term default." The company framed the step as short-term while it works with the Administration on a cyber Executive Order framework and a repeatable process for future releases.

Why it matters

GPT-5.6 bundles technical changes, explicit pricing, and a highly visible safety review into a single launch. The pricing point for Sol, $5 input and $30 output per million tokens, undercuts the cited pricing for Anthropic’s Claude Fable 5 and signals aggressive commercial positioning. The public emphasis on automated red-teaming and the disclosure of approximately 700,000 A100e GPU hours spent on safety testing reflect a heightened operational focus on misuse risk during a period of increased scrutiny from the US government.

What to watch

Watch for who gains access during the preview when the Trump administration vets customers case-by-case, and watch for OpenAI’s follow-up on general availability in the coming weeks. Also track the findings from the two-week third-party testing window and any details that emerge from the Administration’s work on the planned cyber Executive Order framework.