AI Daily Grind // AI news, freshly ground daily

Learn · Models and how they work

What is Mixture of Experts?

Illustration: Mixture of Experts
Simply put, like I'm 5 years oldImagine a big hospital with many specialists, but for each question you only see the two doctors who know that subject best. The whole hospital is big, but each answer only needs a few of them.

The slightly longer version

An AI model is made of billions of numbers called parameters. They are the settings the model learned during training. Usually, every parameter works on every word the model reads or writes. In a mixture of experts design, the model is cut into many smaller parts called experts. A small piece called a router looks at each word and sends it to only a few experts.

That is why people say Flint has 40 billion parameters but only 2.5 billion active at a time. The model knows a lot, but each step does much less math. Less math means faster answers on a normal laptop. One catch: the whole model usually still needs to fit in the computer's memory, even if most of it rests at any moment.

How a mixture of experts answers

  1. Words come inYou type a question or some code.
  2. Router decidesA small part of the model picks which experts fit each word.
  3. Few experts workOnly those chosen experts do the math, the others rest.
  4. Results combinedTheir outputs are mixed together into the next word.
  5. Fast answerLess math per word means quicker answers on normal hardware.
Why it matters

This design is a big reason models like Flint can run on a laptop, for free, without sending your code to a cloud company. That means more privacy and less dependence on a few big labs. For a developer who cannot pay for big cloud models, this is good news.

What to watch out for

Speed is not the same as quality: with fewer active parameters, a model can still miss things a bigger model would catch. Flint's claim that it beats Qwen3.6-35B comes from the company's own tables, and users are still testing it, so wait for independent reviews. Also, the 40 billion and 2.5 billion numbers come from users on a Reddit forum, not from a full official report.

Where it showed up

This notion came up in AI finds the holes, AI picks the locks, and the agents want your wallet. More lessons: Learn.

Sources

Written on 2026-10-29. This explanation was accurate on that day. AI changes fast, so check the sources and the date before you rely on it.

Get The AI Daily Grind by email, every morning, in simple words.