OpenAI ships GPT-5.6 Sol, Terra and Luna to ~20 vetted partners after Trump executive order
The flagship Sol model, a new 'ultra' multi-agent mode and a tiered Terra/Luna lineup launched Friday under a White House-coordinated limited preview that OpenAI says should not become the long-term default.
OpenAI shipped GPT-5.6 on Friday to roughly 20 vetted partner organizations under a White House-coordinated limited preview, the first frontier release to flow through the federal benchmarking process President Trump established by executive order on June 2, 2026. The company didn’t name the partners; Reuters reports it declined to.
The lineup is tiered. Sol is the flagship, priced at $5 per million input tokens and $30 per million output tokens, identical to GPT-5.5 on the input side. Terra lands at half of Sol, or roughly $2.50 input and $15 output. Luna, the budget tier, comes in at $1 input and $6 output. A new “ultra” multi-agent mode and a “max” reasoning setting ship alongside.
OpenAI calls the restricted rollout a “short-term step” and says broader availability is expected in the coming weeks. The blog post is explicit that the company doesn’t believe “this kind of government access process should become the long-term default,” while also describing the arrangement as a “cyber Executive Order framework and a repeatable process for future model releases.” Both things, apparently, at once.
The benchmark story is the case for taking Sol seriously. It sets a state of the art on Terminal-Bench 2.1, clears 50.9% on Agent’s Last Exam in code mode, and matched Anthropic’s Mythos Preview on ExploitBench using roughly one-third the output tokens. OpenAI says it spent over 700,000 A100-equivalent GPU hours on automated red teaming for Sol alone. A Cerebras partnership targeting up to 750 tokens per second begins in July 2026.
The safety story complicates the access story. Per the system card, all three models are rated “High” on Cybersecurity and Biological and Chemical risk under the Preparedness Framework; none clears the High threshold for AI Self-Improvement. Sol and Terra can locate vulnerabilities and assemble exploit fragments but can’t run autonomous end-to-end attacks against hardened targets. That capability profile is precisely what the executive order was written to gate.
Dean Ball, former White House AI adviser and an incoming OpenAI employee, called the June 2 order a “de facto involuntary licensing regime” for frontier AI. The most interesting fact about that quote is who’s saying it, and where he’s about to work.
Sources
- https://openai.com/index/previewing-gpt-5-6-sol/
- https://deploymentsafety.openai.com/gpt-5-6-preview
- https://techcrunch.com/2026/06/26/openai-limits-gpt-5-6-rollout-after-government-request-says-restrictions-shouldnt-be-the-norm/
- https://www.reuters.com/technology/openai-defers-public-rollout-gpt-5-6-us-seeks-early-access-frontier-ai-models-2026-06-26/
- 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
— END —