By Microsoft shipped two on-device AI models at Build 2026 today — Aion 1.0 Instruct and Aion 1.0 Plan — and for the first time, a reasoning model capable of orchestrating sub-agents comes built directly into Windows. No API key. No cloud round trip. No token bill. The OS just became an AI runtime, and it changes how you should be thinking about your next app’s architecture.
Two Models, Two Jobs
Microsoft is drawing a clean line between the two Aion models. Aion 1.0 Instruct is the everyday workhorse: a small language model (SLM) that handles summarization, text rewriting, intent detection, and accessibility features. It replaces Phi Silica as the Windows on-device SLM, and it’s faster and more efficient than its predecessor. More importantly, it runs on CPU — no NPU, no dedicated GPU required — which means it works on far more PCs than any prior Windows AI model.
Aion 1.0 Plan is the bigger bet. It’s a 14-billion parameter reasoning and tool-calling model with a 32K context window. It ships in-box with Windows on capable devices and enables applications to reason over user intent, invoke tools, manage files, and orchestrate sub-agents — all without a cloud connection. Microsoft calls it “fully agentic workflows on the device.” That’s not marketing copy. A 14B model with tool-calling is a real reasoning system. This is meaningful infrastructure.
Try Aion 1.0 Instruct in Edge Canary Today
Aion 1.0 Plan isn’t available yet (arriving “in the coming months”), but Aion 1.0 Instruct is accessible right now through Edge Canary. Here’s how:
- Download Edge Canary or Dev (version 150.0.4070 or later)
- Navigate to
edge://flags - Search for Enable prerelease on-device language model
- Set the dropdown to Enabled and restart Edge
- Confirm at
edge://on-device-internals→ Model Status → Model Name should read Aion-1.0-Instruct
The model downloads automatically on first use. Once enabled, you can call it from JavaScript using Edge’s built-in AI APIs.
The Web Developer Angle: AI Without a Backend
This is where things get interesting for web developers. Edge’s Prompt API, Summarizer, Writer, and Rewriter APIs now run on top of Aion 1.0 Instruct. A few lines of JavaScript is all it takes:
// Summarize page content on-device — no server, no API key
const summarizer = await window.ai.summarizer.create();
const summary = await summarizer.summarize(document.body.innerText);
console.log(summary);No backend server. No API costs. User data stays on the device. This directly competes with Chrome’s Prompt API (backed by Gemini Nano), and Microsoft has a card Chrome doesn’t: Aion 1.0 Plan’s eventual tool-calling capabilities will go far beyond what Gemini Nano can do.
Why CPU Inference Is the Underrated Detail
The specification that matters most here isn’t the 14B parameter count — it’s that Aion 1.0 Instruct runs on CPU. Every other on-device AI story of 2026 has a hardware asterisk: Apple Foundation Models require Apple Silicon, many Windows AI features require a Copilot+ PC with an NPU. Aion 1.0 Instruct has no such restriction. If your user is on Windows 11 with a modern CPU, your app can deliver on-device AI. That dramatically expands the addressable developer market.
Microsoft Is Building Its Own AI Stack
Read the subtext here. Aion models are built by Microsoft Research, not OpenAI. CNBC reported it plainly: Microsoft is reducing its reliance on OpenAI and lowering costs for developers. Every inference that runs on Aion is one that doesn’t route through an OpenAI API call — saving Microsoft money and giving it more control over its platform. For developers, this means the Windows AI stack you build on is Microsoft’s independent investment, not a repackaged OpenAI product.
What Happens Next
Two dates to track:
- July 2026: Aion 1.0 Instruct open weights land on Hugging Face. You’ll be able to fine-tune it for specialized use cases and run it via Windows AI APIs.
- Coming months: Aion 1.0 Plan reaches general availability — full agentic workflows running locally on Windows, with no cloud dependency required.
For now: download Edge Canary, flip the flag, and start prototyping with Aion 1.0 Instruct. The architecture decisions you make today — cloud-first, on-device, or hybrid — will be far easier to revisit once you’ve seen what local inference actually feels like at this quality level. Bookmark the Windows Developer Blog for the Aion 1.0 Plan GA announcement.
