Aion Instruct at a glance
Microsoft announced Aion Instruct on November 13, 2026: a 2.3B-parameter on-device Windows language model with a 32K-token input context, replacing Phi Silica (3.8B, 3.5K). A test package is available now; Insider rollout starts November 20 and retail is targeted for late January 2027.
Ask your AI about this story
Use this story in your business: Microsoft Aion Instruct: 2.3B on-device Windows model with 32K context replaces Phi Silica, test package available now (Microsoft). Ask your AI to explain what happened, how it could affect your business and how to put it to work. It also weighs price, speed, intelligence and the lab behind it. The link to the story is included.
What changed with Aion Instruct
Microsoft's Foundry on Windows post of November 13, 2026 sets out the transition from Phi Silica to Aion Instruct, its newest on-device language model for Windows. Aion Instruct has 2.3 billion parameters against Phi Silica's 3.8 billion, and raises the input context limit from 3.5K to 32K tokens. Microsoft is also expanding it beyond Copilot+ PCs to supported GPUs, including NVIDIA RTX-series and AMD RDNA4 cards, and has added invisible watermarking to meet text-provenance requirements in specific regions. A standalone test package has been available since October 2026 at aka.ms/tryaion, so developers can validate their apps end to end today.
The staged rollout begins on Windows Insider devices on November 20, 2026, with full retail availability targeted for late January 2027, when Microsoft also targets removing the LAF token requirement. Most apps need no action, but apps using highly custom prompting, custom LoRAs or custom output parsing must be re-validated, because the new model's behavior and output strings differ. As of November 14, 2026, Artificial Analysis had not published a page for Aion Instruct, so this item carries no independent benchmark metrics, and Microsoft published no price (the model ships with Windows).
Aion Instruct in numbers
Values are copied from the linked source on the date shown. They are not IA Compa measurements.
- Input context windowas of · Source for Input context window: devblogs.microsoft.com (opens in a new tab)
- 32,000
How it compares
Current models we have covered: released in the last six months, newest version of each line only. Each value is the model's latest recorded value. (Models)
Where to compare Aion Instruct
Independent leaderboards and official resources for comparing this model yourself. Rankings change often; each site shows its own current data.
General leaderboards
Why Aion Instruct matters for businesses
Why it matters: every Windows PC your business runs is about to get a smaller, longer-context local model underneath its built-in AI features, with no cloud bill and no data leaving the device. If your team sells or supports Windows apps that use the local model — summarization, drafting, on-device assistants — budget time before January to re-test any custom prompts or LoRAs against the test package, because the January switch happens behind the scenes and silently changes model behavior.
Put it to work: AI services
Get AI News by email
Verified AI stories, each with its official source, on Monday and Thursday mornings. Add breaking news to hear about the biggest ones as soon as we publish them.
Every story links its official source
Free, in English or Spanish
Unsubscribe with one click, any time
Sources for this story
- Microsoft Aion Instruct: Advancing local AI on Windows and expanding to GPU (opens in a new tab)Official
Microsoft · devblogs.microsoft.com
Last verified: . IA Compa summarizes public announcements for business readers. We are not affiliated with the companies mentioned. Always confirm details, pricing, and availability with the official source.
- #aion-instruct
- #microsoft
- #on-device-ai
- #windows
- #phi-silica
- #local-models
- #gpu

