Mistral 3: Major New Vision Model Release

Mistral 3: Major New Vision Model Release

🚀 Introducing Mistral 3

Mistral has just launched four new AI models, expanding their multimodal capabilities:

  • Ministral 14B — Compact, high-performance dense model
  • Ministral 8B — Balanced efficiency and scale
  • Ministral 3B — Ultra-lightweight vision model (~3GB)
  • Mistral Large 3 MoEMixture of Experts model with 675B parameters (41B active at runtime)

All models are vision-capable and released under the Apache 2.0 open-source license.

---

✨ Highlight: Ministral 3B — AI in Your Browser

The 3B model is especially intriguing thanks to its small size and strong multimodal performance. Hugging Face developer Xenova even ran it directly in a browser using WebGPU — no server required:

> The 3B model is small enough to run entirely locally in your browser using WebGPU! 🤯

Try it yourself:

  • WebGPU Browser Demo
  • Loads the 3GB model in-browser
  • Streams input directly from your webcam
  • Accepts text prompts based on live visual input
  • No external API calls or cloud services involved
image

---

🔧 Using Mistral 3 with the `llm-mistral` Plugin

Mistral’s API-hosted models are already supported by my `llm-mistral` plugin.

Simply run:

$ llm mistral refresh
Added models: ministral-3b-2512, ministral-14b-latest, mistral-large-2512, ministral-14b-2512, ministral-8b-2512

This updates your local environment with the latest available models.

---

🖼 Model Performance: "Pelican" Test

A quick benchmark with an identical "pelican" prompt produced the following results:

Best output — Mistral Large 3:

image

Least impressive — Ministral 3B:

image

---

🌍 Why This Matters: Local & Cross-Platform AI for Creators

As open-source AI models like Mistral become:

  • Smaller
  • More capable
  • Cross-platform compatible

…creators can run them entirely locally or deploy via APIs, enabling powerful new workflows.

One example is AiToEarn:

  • Helps creators deploy AI models for content creation, publishing, and monetization
  • Works across Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X
  • Includes an open-source toolkit offering:
  • AI generation
  • Cross-platform publishing
  • Analytics
  • Model performance rankings
  • Supports both local and hosted AI deployment for maximum profitability and efficiency

---

Bottom line:

Mistral 3 marks a turning point. Powerful, vision-enabled AI models can now run entirely in your browser or integrate easily into creative workflows — all while remaining fully open-source.

Read more

Translate the following blog post title into English, concise and natural. Return plain text only without quotes. 哈佛大学 R 编程课程介绍

Harvard CS50: Introduction to Programming with R Harvard University offers exceptional beginner-friendly computer science courses. We’re excited to announce the release of Harvard CS50’s Introduction to Programming in R, a powerful language widely used for statistical computing, data science, and graphics. This course was developed by Carter Zenke.