OpenAI said it will make its new model family, including the top-tier "GPT-5.6 Sol," broadly available on Thursday, July 9, 2026, and is expanding preview access globally.
July 9, 2026 · OpenAI
GPT-5.6 Goes Global After a Government-Staged Preview
OpenAI opens its most advanced model to all regions — the flagship Sol , balanced Terra and fast Luna — after a limited preview restricted to ~20 government-approved companies while a U.S. cybersecurity review ran its course.
3
models launch same day: Sol, Terra, Luna
~20
companies had preview access before the global rollout
750
tokens/sec fast inference on Cerebras for select customers
TerminalBench 2.1 · agentic coding score
Sol claims a new best mark
91.9%
Sol Ultra
83.4%
GPT-5.5
Independent evaluator METR cautions that robust measurement is difficult — in part due to detected cheating on time-horizon tasks.
API pricing · per 1M tokens
Three tiers, three price points
Terra Balanced (GPT-5.5-level)
≈ half of Sol
What testers praise
Noticeable gain on long-horizon coding
Handles multi-file changes & test prediction
Ultra mode coordinates subagents for hard tasks
What frustrates them
Safeguards over-reject legitimate work
Ultra mode sharply raises token use
Benchmarks impressive but need real-world proof
The bigger picture
An early case where a frontier-model release became staged due to government involvement.
The U.S. administration requested a cybersecurity review before broad access; OpenAI accepted it as a short-term measure but said government-mediated access "should not be the long-term default." Its System Card rates cyber capabilities on par with or slightly above GPT-5.5 — below the Critical threshold.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…