OpenAI is rolling out a GPT-5.6 preview featuring three new models—Sol, Terra, and Luna—over the coming weeks. Sol is positioned as a flagship for cybersecurity, coding, and biological sciences, with upgraded performance and lower token use, plus new “max” and
By the time OpenAI finishes rolling out its GPT-5.6 preview, it wants users to feel a clear difference in both capability and control. This launch isn’t just one model—it’s a trio: Sol, Terra, and Luna—arriving over the coming weeks.
Sol is the headline act. OpenAI frames GPT-5.6 Sol as its new flagship, optimized for performance across cybersecurity, biological sciences, and general coding tasks. It’s also positioned as an upgrade over GPT-5.5 across many workflows. with efficiency improvements that are expected to cut token consumption.
OpenAI is also adding new ways to push the model harder. Sol comes with two new modes—“max” and “ultra”—designed to change how deeply it reasons. “Max” is described as enabling deeper reasoning, while “ultra” is tied to leveraging multiple agents.
Terra takes a different route. It’s designed to be “just right,” balancing performance and speed. OpenAI says Terra will operate much like the GPT-5.5 users have now, but at less than half the expense.
Then there’s Luna, whose pitch is blunt: efficiency first. OpenAI characterizes GPT-5.6 Luna as delivering “strong capability,” but with pricing over 50% lower than Terra’s.
All of that performance talk runs right into the “national security elephant in the room,” and OpenAI addresses it directly: Sol, Terra, and Luna are described as being equipped with multiple levels of safeguards intended to strengthen them against being used for nefarious purposes.
For Sol in particular, OpenAI says it’s been tuned to excel at finding software vulnerabilities and developing fixes—while resisting efforts to craft full exploit chains ready for attackers to use.
The company also acknowledges a reality that’s become familiar to anyone who’s watched AI safety barriers collide with real-world use. OpenAI warns that during this initial GPT-5.6 preview. those safeguards might be erring on the side of caution. preventing users from performing legitimate tasks. OpenAI expects to further fine-tune the approach, with the goal of making the process less burdensome over time.
The preview itself will start narrow. OpenAI is making the GPT-5.6 preview available to “trusted partners and organizations,” with broader availability—including in ChatGPT and Codex—planned for later.
OpenAI GPT-5.6 Sol Terra Luna cybersecurity AI models token consumption safeguards ChatGPT Codex trusted partners
So it’s basically safer hacking but still for hacking. Got it.
Lower token use?? Cool cool. But if they’re “erring on the side of caution” then I’m sure it’ll block regular stuff too, like trying to code something simple.
“Ultra” sounds like multiple agents doing the thinking, which is just marketing for more loops. Also why is it called Sol/ Terra/ Luna like it’s a space program? This whole time I thought OpenAI was already GPT-5.5 so what is this extra step supposed to do.
Trusted partners only at first… so yeah, it’ll be nerfed for normal people then “rebalanced” later. I swear these safeguards always end up stopping the good uses like helping small businesses write scripts, but attackers will still find a way. Also “resist exploit chains” like they’re gonna monitor what you type forever? Sounds good on paper though.