Magistral Small 1.2 brings a 24B open-source reasoning model to local PCs

In short: Magistral Small 1.2 is a new open-source AI model focused on reasoning, and it is designed to run on high-end consumer devices.

What happened

Magistral Small 1.2 was released in September 2025 by the Magistral model family developers. It is described as a 24 billion parameter AI model, meaning it has 24 billion adjustable “knobs” inside it that help it learn patterns. That is smaller than the biggest AI models, but still large enough to handle complex tasks.

The developers say it can run locally on consumer hardware, like a single Nvidia RTX 4090 graphics card or a 32GB MacBook, when it is “quantized” (a way of compressing the model so it uses less memory, like saving a photo at a smaller file size). It also supports multimodality, which means it can work with more than one type of input, like text and images.

Magistral Small 1.2 includes “visible chain-of-thought,” which is the model showing the steps it took to reach an answer. The goal is auditability, especially in areas like legal work, finance, and healthcare where people may need to review how a conclusion was reached. It also has a 128,000 token context window, which is the amount of text it can keep in mind at once (like having a much longer short-term memory).

The model is licensed under Apache 2.0, a permissive license that allows broad reuse. The same family also offers an API with pricing listed at $0.50 and $1.50, depending on usage.

One note of confusion is that some descriptions mention a 2 billion parameter model with 14 languages. The available details for Magistral Small 1.2 point to 24 billion parameters, and language support is not confirmed as “14” in the provided information.

Why it matters

Running a capable AI model on your own computer can help with privacy and cost, since data does not need to be sent to a remote server for every request.

Source: TechCrunch AI