OpenAI GPT-5.6 Sol launches in restricted US-only preview

GPT‑5.6 is significant not just because it’s OpenAI’s new flagship, but because it lands squarely at the intersection of technical progress and state control. On paper, Sol, Terra and Luna are a straightforward cadence upgrade: better long‑horizon reasoning, stronger coding and biology tools, and a big step up in cyber capabilities. Benchmarks like Terminal‑Bench and ExploitBench suggest Sol is now competitive with—or slightly ahead of—Anthropic’s Mythos‑class systems on high‑stakes security tasks while using fewer tokens, a telling data point in the frontier model arms race.

What really matters for the AGI race, though, is the launch pattern. For the first time, a top‑tier frontier model is being rolled out under an explicit, temporary gate run by the US government, with OpenAI handing over early access and detailed evaluations before a wider public release. That sets a powerful precedent: the bottleneck for frontier capability may increasingly be regulatory throughput, not just compute or research talent. For rivals, GPT‑5.6 raises the bar on agentic, tool‑using systems—especially for coding and security—while also normalizing a world where national security reviews are just part of shipping a new model.