Big AI model announcements now come with two size numbers, like “816B total, 42B active.” For a long time there was just one number. The second one shows up because of how a lot of the biggest models are built now. If you’ve seen a model name like “235B-A22B,” that’s both numbers jammed into the title.
Both numbers are real. They just measure different things, and companies don’t always tell you which one they’re bragging about.
Total Parameters
Parameters are the adjustable numbers a model learns during training. Total parameters is all of them, everything the model learned, sitting in storage. And all of it has to be loaded into memory to run the model at all, which is why the total number decides what hardware you need.
Meta’s Muse Spark Open, released October 15, 2026, has 816 billion total. Its model card says that takes at least 700GB of GPU memory, even in a compressed format. That’s well beyond a gaming PC (Meta does say it’ll run on two top-end Mac Studios, which is still a lot of Mac).
Active Parameters
Active parameters are the ones that actually do the work for each token (chunk of text) the model writes. In a Mixture of Experts model, the network is split into lots of smaller “experts,” and a router picks just a few of them for each token.
Muse Spark Open has 256 experts and uses 8 per token, so only 42 billion of its 816 billion parameters fire at once. That’s why it’s faster and cheaper per answer than its total size suggests. DeepSeek-V3 does the same trick: 671 billion total, 37 billion active. It’s also why two models with the same total size can feel really different in speed and cost when you actually use them.
The tell
Total is about storage. Active is about effort. Total decides how big a computer you need to run the model at all. Active decides how fast and cheap each answer is.
A hospital has 300 doctors on staff, but your visit only needs the four who handle your problem. The staff list is the total. Your appointment is the active part. You still need a building big enough for all 300.
So if someone quotes one number, ask which one. A “42B model” and an “816B model” can be the exact same model, depending on who’s trying to impress you.

