---
title: Choose an open-source LLM
description: Teach Anything runs on open-source LLMs. Here is what each one is good at, so you can pick the model that fits your subject.
sidebar:
  label: Choose an LLM
  icon: cpu
---

Teach Anything runs entirely on **open-source large language models** — the "engines" that write your assistant's answers. You are never locked into one vendor's model, and you can switch between them at any time from your assistant's settings.

Each model has its own strengths. Depending on your subject matter, you can choose a model that is best tailored to your domain and goals.

## The models we offer

**[GPT-OSS 120B (OpenAI)](https://openai.com/index/introducing-gpt-oss/)**

Strong general reasoning and good at following instructions.

**[Llama 4 Maverick (Meta)](https://ai.meta.com/blog/llama-4-multimodal-intelligence/)**

Huge context window — best for very long documents.

**[Qwen 3 235B (Alibaba's Qwen team)](https://qwen.ai/blog?id=qwen3)**

Suitable for STEM subjects and math, with strong multilingual support. A
good pick for courses taught in a language other than English.

**[Llama 3.3 70B (Meta)](https://developer.meta.com/ai/docs/model-cards-and-prompt-formats/llama3_3/)**

A dependable all-rounder. This is our default model.

**[DeepSeek V3.2 (DeepSeek)](https://api-docs.deepseek.com/news/news251201/)**

Excels at complex reasoning and coding.

Each card links to the model maker's own announcement if you want to read more.

## Picking one

Give each model a try. You can easily switch among them at any time, and nothing else about your assistant changes when you do — your materials, instructions, and conversations all stay put.

1. **Open your assistant's settings**

    From your [dashboard](https://teachanything.ai/dashboard), open the
    assistant you want to change.

2. **Pick a model from the LLM list**

    Models are grouped by the organization that built them.

3. **Ask it a few of your course's typical questions**

    Compare the answers side by side. If the new model isn't an improvement,
    switch back — the change takes effect right away.

:::tip[When in doubt, stay on the default]
Llama 3.3 70B is a well-rounded choice that works for most courses. Reach for a
different model when you have a specific reason: teaching in a language other
than English, working with very long documents, or covering heavy math or code.
:::

## Going deeper

If you're interested, you can explore independent [technical analyses of open-source LLMs](https://artificialanalysis.ai/models/open-source), which benchmark these models side by side on quality, speed, and cost.

Prof. Joubin also writes about model choice on her Substack: [More open-source LLMs to choose from](https://ajoubin.substack.com/p/more-open-source-llms-to-choose-from).

Teach Anything will continue to expand and update our offering of open-source LLMs.

**[Next: Customize your assistant](/instructors/customize)**

Tune the instructions, temperature, answer length, and source citations.