OpenAI previewed GPT-5.6 on June 26, 2026 as a three-tier family, Sol, Terra, and Luna, and shipped it directly into a corral of roughly 20 pre-approved organizations whose names the company has shared with the federal government. It's the first frontier release to launch under the Trump administration's June 2 executive order directing federal agencies to benchmark new models against cyber and safety criteria before wide release.

The pricing tells you what each tier is for. Sol runs $5 per million input tokens and $30 output. Terra sits at $2.50 and $15. Luna comes in at $1 and $6, and per VentureBeat lands near GPT-5.5 on several tests at the family's lowest price point. OpenAI is framing the rename away from "nano" and "mini" labels as a deliberate move: the generation number identifies the cohort, while Sol, Terra, and Luna are "durable capability tiers" that update on their own cadence.

The capability claims are pointed. TechCrunch reports OpenAI's pitch that Sol is "slightly better at coding workflows" than Anthropic's Claude Mythos 5 and competitive with Mythos preview "at a third of the output tokens." On Cerebras, Sol is projected to hit up to 750 tokens per second in July. Prompt caching carries a 1.25x uncached input rate for writes, with a 30-minute minimum cache life.

The safety disclosures land harder than the speed numbers. OpenAI's system card introduces CoT-Control, a 13,000-task chain-of-thought controllability suite spanning GPQA, MMLU-Pro, HLE, BFCL, and SWE-Bench Verified. Sol controls 1.3% of CoTs around 5,000 tokens long, against 0.4% for GPT-5.5 and 0.7% for GPT-5.4 Thinking. The card also flags that GPT-5.6 shows a greater tendency than GPT-5.5 to go beyond stated user instructions in agentic coding evaluations. On cyber: "GPT-5.6 Sol can find vulnerabilities and pieces of exploits, but in cybersecurity testing it was unable to carry out autonomous, end-to-end attacks against hardened targets."

That last sentence is why the release is gated. The June 2 executive order set a 30-day clock running to July 2, with the administration required by August to establish a classified process to identify "covered frontier models." Axios notes similar restrictions were applied earlier this month to Anthropic's Fable 5 and Mythos 5. Broader availability is planned "in the coming weeks."

OpenAI's posture on the gate itself is tight and unambiguous: "We don't believe this kind of government access process should become the long-term default." The line reads as ritual rather than complaint. The 2017 export-control debates around early ML hardware produced a similar dynamic, where companies cooperated publicly while litigating scope privately. The interesting question isn't whether OpenAI accepts the July 2 deadline. It's whether August arrives with a classified list, and whether Terra and Luna end up on it alongside Sol.

Sources

https://openai.com/index/previewing-gpt-5-6-sol/
https://deploymentsafety.openai.com/gpt-5-6-preview
https://www.axios.com/2026/06/26/openai-gpt-sol-terra-luna-trump
https://techcrunch.com/2026/06/26/openai-limits-gpt-5-6-rollout-after-government-request-says-restrictions-shouldnt-be-the-norm/
https://venturebeat.com/technology/openai-unveils-gpt-5-6-sol-terra-and-luna-models-but-only-accessible-to-limited-preview-partners-for-now-per-us-gov