The AI industry has a vocabulary problem. Every new model is "powerful," "intelligent," "groundbreaking," approaching something that sounds like a mind. The language has inflated so far that the words have stopped meaning anything. So we have made a deliberate choice to do the opposite: we will tell you exactly what our models do, exactly what they don't, and we will not call any of it "smart."
This is not modesty. It is the only honest way to talk about a tool, and we think honesty is the most valuable thing a research company can offer in a field that has run out of it.
When a company calls its model "intelligent," it is usually hoping you won't ask the follow-up questions. Intelligent at what? Measured how? Compared to what? Does it still work on a problem it has never seen, or only on the ones it was quietly trained to handle?
"Smart" is a word that ends conversations. A real claim invites them. Instead of saying our model is smart, we will say: it answers reasoning problems harder than the ones it trained on, at this measured accuracy, against this specific baseline, reproducible from this exact script. That is a sentence you can check. "Smart" is a sentence you can only believe.
A claim you can verify is worth more than a claim you have to trust.
This honesty is not just marketing posture; it shapes how we work. We test every capability on problems the model has never seen. We keep our failures and publish them alongside the wins. We tie every number to a script anyone can re-run. And when a result doesn't hold up, we say so plainly rather than burying it.
It would be easier to do the opposite. "Smart" sells. Carefully-bounded claims do not make headlines. But a headline that gets debunked is worth less than no headline at all, and the moment you oversell once, every honest claim you make afterward is in doubt.
So here is the deal we are offering. We will never tell you our model is smarter than it is. We will show you what it does, how we measured it, and where it falls short. When we are uncertain, we will say "we don't know yet" instead of dressing up a guess. And when we eventually build something genuinely impressive, you will be able to trust it — precisely because we spent all the time before it refusing to exaggerate.
In a field full of models that claim to think, we would rather be the one that tells you the truth.