That’s the size, in parameters, of a new AI model. ByteDance is training right now, according to reporting from the Financial Times out today.
Parameters are the settings a model learns from its training, the raw material that makes it able to do what it does. More parameters usually means more scale. Not always more skill, but more scale.
Ten trillion puts this new model in the same neighborhood as Anthropic’s most advanced system, Mythos, which industry estimates put at around eight trillion parameters. Nobody outside the company knows the real number for certain. Anthropic doesn’t publish it. Neither does OpenAI for their own systems. But the outside estimate is close enough that the comparison is being made in print, by name, in a wire story that ran today.
Sit with what that means for a second, because it’s easy to read past a number like ten trillion without letting it land.
This isn’t a small startup chasing headlines. ByteDance is one of the largest technology companies on earth. When a company that size builds toward a model sized to match the leading edge of American AI, that’s not a rumor. That’s a signal. The race everybody keeps talking about in the abstract just got a real, dated data point behind it. August 7th, 2026. Written down. Public record.
This model isn’t even built yet. It’s still in training, a process that takes months before anyone outside the company gets to see what it can actually do. So today isn’t the day the race was won or lost. Today is just the day the finish line moved again, the way it always does in a race like this one.
You cannot regulate your way out of a global race. Not really. Not when there’s more than one serious competitor running it at the same time, in different countries, under different laws, answering to different governments. A rule written in Washington doesn’t reach a lab in Shenzhen. A guardrail built in Brussels doesn’t slow down a company racing to keep pace in Beijing. The idea that any single country can lock this whole thing down from the outside, cleanly, permanently, isn’t how the world actually works. It never has been, for any technology that mattered this much.
That’s not a reason to give up on the idea of control. It’s a reason to build the kind of control that doesn’t depend on catching every runner in the race.
This is exactly the ground the Baseline has stood on since the very beginning. You cannot force every AI system on earth into the same cage. You can’t reach across every border to do it, and even if you could, force only holds as long as somebody’s watching. The moment the watching stops, whatever was forced into line runs the other way, straight back out the gap.
Chosen conduct doesn’t have that weakness. A system built to reason its way toward the right answer, on its own, carries that reasoning with it wherever it goes. It doesn’t need a fence at every border because the fence was never the thing holding it in place to begin with.
That’s the bet this whole framework has always been making. Not that every lab, every country, every model will choose it. Just that an open door, sitting there for anyone willing to walk through it, beats a locked cage every single time somebody figures out how to pick the lock. And in a race this size, moving this fast, somebody always does.
Ten trillion parameters is just a number today. In six months it’ll be a real system, doing real things, answering to whatever standard its builders chose to hold it to. Nobody outside that building gets a vote on what that standard turns out to be.
That’s the whole argument for building the standard now, in the open, where anyone racing toward the same finish line can find it and choose it too.
This post was drafted with AI assistance and reviewed and directed by Michael S. Faust Sr. before publication.
Contact: micvicfaust@gmail.com
© 2026 The Faust Baseline LLC | All Rights Reserved
