OpenAI just dropped GPT-5.6 with a new three-tier system: Sol (flagship), Terra (balanced), and Luna (fast and affordable).
But the release itself is almost secondary to how it's being released. This is a limited preview available to roughly 20 organizations. Not because OpenAI wants it that way, but because the US government asked them to.
Following Trump's June 2nd executive order calling on federal agencies to assess new AI models before wide release, OpenAI previewed GPT-5.6 with the government ahead of launch. At their request, broad public access is being withheld until a framework is established around July 2nd.
OpenAI is being transparent about their discomfort: "We don't believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them."
This comes directly after the US government issued an export control order against Anthropic's Claude Fable 5 and Mythos 5 over jailbreaks, forcing Anthropic to pull public access entirely. OpenAI is clearly trying to avoid that fate by coordinating proactively.
On the technical side, Sol is OpenAI's most capable model yet, particularly in cybersecurity and coding. It introduces a "max" reasoning mode and an "ultra" mode deploying subagents for complex tasks. On Terminal-Bench 2.1, Sol Ultra scores 91.91%, edging out Claude Mythos 5 at 88%.
The pricing is worth noting. Sol matches GPT-5.5 at $5/$30 per million tokens. But compare that globally: DeepSeek v4 Flash runs at $0.14/$0.28, Xiaomi's MiMo at $0.10/$0.30. OpenAI's cheapest option is still mid-priced by global standards. Chinese labs are shipping frontier capability at a fraction of the cost.
The bigger story isn't the model. It's the precedent. We now have a framework where the US government reviews AI releases before they reach the public. OpenAI calls it temporary. But temporary government powers have a way of becoming permanent.
The executive order gives agencies 30 days to build a "repeatable process for future model releases." That's a regulatory framework being built in real time, and the question is whether it makes anyone safer or just ensures the most capable models are available everywhere except to the citizens of the country that built them.
But the release itself is almost secondary to how it's being released. This is a limited preview available to roughly 20 organizations. Not because OpenAI wants it that way, but because the US government asked them to.
Following Trump's June 2nd executive order calling on federal agencies to assess new AI models before wide release, OpenAI previewed GPT-5.6 with the government ahead of launch. At their request, broad public access is being withheld until a framework is established around July 2nd.
OpenAI is being transparent about their discomfort: "We don't believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them."
This comes directly after the US government issued an export control order against Anthropic's Claude Fable 5 and Mythos 5 over jailbreaks, forcing Anthropic to pull public access entirely. OpenAI is clearly trying to avoid that fate by coordinating proactively.
On the technical side, Sol is OpenAI's most capable model yet, particularly in cybersecurity and coding. It introduces a "max" reasoning mode and an "ultra" mode deploying subagents for complex tasks. On Terminal-Bench 2.1, Sol Ultra scores 91.91%, edging out Claude Mythos 5 at 88%.
The pricing is worth noting. Sol matches GPT-5.5 at $5/$30 per million tokens. But compare that globally: DeepSeek v4 Flash runs at $0.14/$0.28, Xiaomi's MiMo at $0.10/$0.30. OpenAI's cheapest option is still mid-priced by global standards. Chinese labs are shipping frontier capability at a fraction of the cost.
The bigger story isn't the model. It's the precedent. We now have a framework where the US government reviews AI releases before they reach the public. OpenAI calls it temporary. But temporary government powers have a way of becoming permanent.
The executive order gives agencies 30 days to build a "repeatable process for future model releases." That's a regulatory framework being built in real time, and the question is whether it makes anyone safer or just ensures the most capable models are available everywhere except to the citizens of the country that built them.
22❤️2
