Remember when naming an AI model just meant bolting a bigger number onto the box? OpenAI apparently got bored of that and raided a planetarium gift shop instead. Meet Sol, Terra, and Luna — the new GPT-5.6 family — and yes, one of them runs faster than you can finish reading this sentence.
Three Models Walk Into a Benchmark
On June 26, OpenAI launched the GPT-5.6 series with a clear pecking order: Sol as the flagship, Terra as the balanced middle child, and Luna as the fast, affordable one your finance team actually approves of. Sol set a new state of the art on Terminal-Bench 2.1, the benchmark that grades how gracefully a model survives the command line without accidentally deleting your production database. It also posted strong marks on tougher scientific reasoning tests, because apparently topping coding leaderboards wasn't flex enough.
Pricing spans from Luna at a friendly $1 in / $6 out per million tokens up to Sol at $5 in / $30 out — reasonable, given what these things can do. And in a genuine flex, OpenAI is putting Sol on Cerebras hardware at up to 750 tokens per second starting in July, a speed that makes the word "loading" feel like a quaint historical artifact.
OpenAI also quietly shipped Codex Remote to general availability on June 25, letting you kick off or resume coding work on a connected Mac or Windows host straight from your phone. Because nothing says "work-life balance" like approving a refactor from a beach chair.
Smart Enough to Behave Itself
For the security crowd, the headline is that Sol got noticeably sharper at cybersecurity tasks, identifying vulnerabilities and exploitation components while burning roughly a third fewer output tokens than rival systems. Efficiency and menace, finally shipping in the same package.
But OpenAI is careful to note that Sol did not cross its "Cyber Critical" threshold. In testing it could spot the individual pieces of an exploit yet couldn't autonomously chain them into a functional full attack against hardened browsers. So it's a wildly talented apprentice, not a fully-fledged supervillain — which is precisely where the rest of us would like it to stay, thanks.
Naming your models after celestial bodies is a bold branding swing, but at 750 tokens a second, Sol has earned the right to shine. Just don't ask it to assemble a full exploit chain — it's on its best behavior, and so should you be.