Microsoft unveils Aion 1.0 models: on-device SLM and 14B agentic reasoning

Microsoft unveiled two new on-device AI models at Microsoft Build 2026: Aion 1.0 Instruct and Aion 1.0 Plan, designed to bring powerful AI capabilities directly to Windows devices without cloud dependency. Aion 1.0 Instruct is a next-generation small language model (SLM) optimized for efficiency and speed, targeting everyday text intelligence tasks such as summarization, rewriting, intent recognition, and accessibility features. It is positioned as a direct competitor to Apple's AFM-3B on-device LLM and will be released as open weights, allowing developers to customize and deploy it freely. The model integrates deeply with Windows APIs and the Edge browser, making it a versatile tool for local AI workloads.

Aion 1.0 Plan takes a different approach as a 14-billion parameter reasoning model with a 32,000 token context window. Designed for local agentic reasoning, it enables applications to understand user intent, invoke tools, manage files, and orchestrate sub-agents—all on the device. This brings sophisticated AI workflows to Windows without relying on cloud servers. The Reddit community has speculated whether Plan is simply a fine-tuned version of Microsoft's Phi-4 with reinforcement learning from verified reasoning (RLVR) for tool use, or an entirely new architecture. Microsoft has not confirmed details, but the model's focus on local autonomy marks a significant step toward practical on-device AI agents.