Aion Instruct 2.3B: Microsoft's open MIT-licensed SLM, now Edge's built-in model
Microsoft released Aion-1.0-Instruct-2.3B, a 2.3B-parameter MIT-licensed small language model from the Windows Applied Sciences Group for on-device text tasks, now generally available on Hugging Face and as Microsoft Edge's built-in model, replacing Phi-4-mini.
Why it matters
The model that will become the Windows OS small language model by end of January 2027, open under MIT, running summarization, rewriting and chat entirely on-device with a 32K context.
What you could build with it
An indie developer could build a browser extension that summarizes long web articles and rewrites selected text locally with Aion Instruct through Edge's built-in model, keeping the user's reading private and working offline.
Does it hold up?
Early but usable: the model is GA on Hugging Face and built into Edge, and community GGUF and WebGPU ports appeared within a day, though upstream llama.cpp and Ollama do not load it yet.
Built with Aion Instruct 2.3B
- Aion-1.0-Instruct-2.3B-GGUF: GGUF quants for llama.cpphuggingface · Community GGUF quantizations of Aion 1.0 Instruct with a custom llama.cpp port, since upstream llama.cpp, LM Studio and Ollama do not load the new architecture yet.
- Aion 2.3B WebGPU: Run Microsoft's LLM in the Browserexplainx · Walkthrough of Microsoft's ONNX-WebGPU export of Aion: INT4 weights, 32K context and how to run it in a browser tab via ONNX Runtime GenAI.
- Microsoft Edge AI Update: Aion Instruct Goes GA, Prompt API Trial and Decisions API Previewwindowsforum · Coverage of the GA announcement, the model card's benchmarks (MMLU 55.15, GSM8K 74.91, IFEval 83.90) and the plan for a Windows OS SLM by January 2027.
- microsoft/Aion-Instruct-Preview-Samplegithub · 48 ★ · Microsoft's official sample code for running Aion Instruct on-device, linked from the model card.
Learn more
First spotted on article: source.
More AI models
MiniMind: train your own 64M LLM from scratch in about 2 hoursA complete, from-scratch LLM training pipeline at 64M parameters - pretraining, SFT, LoRA, DPO, PPO/GRPO, tool use and…models · JEV 0.82VoiceStudio: the open-source, fully-local ElevenLabs alternativeA desktop app bundling 20 local speech-generation engines and 10 transcription options behind one UI — voice cloning…models · JEV 0.76Edge0-145B: open 145B MoE that runs on a single consumer GPUEdge0 open-sourced 145B and 120B latent-MoE models derived from Qwen3.8-Flash-Next, plus an inference framework that…models · JEV 0.74Strata: run a 125B model on a gaming PCFree, MIT-licensed local inference engine that runs Qwen3.8-Flash-Next (125B MoE, about 6B active per token) on…models · JEV 0.73tiny-voca: a fully open 4.5M-parameter TTS trained for under $10 of computeA fully open 4.5M-parameter English TTS — code, weights and dataset under Apache 2.0 — trained from scratch in 7.7…models · JEV 0.67Qwen4-Max: Alibaba's new flagship model goes liveAlibaba's 2.4T-parameter flagship is now generally available via API, with a 1M-token context, native vision, and…models · JEV 0.67
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.