OpenAI limits GPT-5.6 rollout at US government request
SkimNews Take
By publicly framing each restriction as an exception rather than codifying policy, OpenAI preserves flexibility to comply with future gating requests case-by-case without committing to a transparent standard.
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI limited its GPT-5.6 rollout to "trusted partners" whose participation has been shared with the U.S. government, restricting all three models in the lineup rather than just the most powerful one
- Trump administration requested the restriction, mirroring its earlier order against Anthropic's Fable 5 — which Anthropic then took down entirely rather than comply with a foreign-national access ban
- Dean Ball, former White House AI adviser and soon-to-be OpenAI employee, called the administration's voluntary pre-release review a "de facto involuntary licensing regime" that could hand China an edge in the AI race without clearly defined safety standards
- OpenAI said in a Friday blog post that the government access process "should not become the long-term default," framing the limited preview as a "short-term step" while it works with the administration on a new executive order framework
- GPT-5.6 Sol adds new "max" and "ultra" reasoning modes for coding, biology, and cybersecurity tasks, with OpenAI claiming it slightly outperforms Anthropic's Claude Mythos 5 on coding benchmarks while using one-third the output tokens
- GPT-5.6 ships with tiered pricing — Luna at $1/$6 per million tokens, Terra at half Sol's price, and Sol at $5/$30 — plus improved prompt caching designed to make repeated queries cheaper and more predictable
Why it matters: For developers and enterprises waiting on OpenAI's most capable models, the immediate consequence is delayed access to Sol's agentic coding and cybersecurity upgrades while OpenAI negotiates a repeatable release framework with the administration. The longer-term stakes Ball flags are concrete: indefinite pre-release reviews without clear safety standards risk delaying launches enough to help Chinese AI labs close the gap and jeopardize billions in planned U.S. AI infrastructure investment.
