The RTX Spark hardware brief in this pipeline covers the chip. Aion 1.0 is what runs on it. Microsoft announced the model family at Build 2026 on June 2, describing two distinct variants for different use cases, and announcing a commitment to open-source the Instruct variant on Hugging Face in July 2026.
All specifications below are vendor-stated from Microsoft’s Build 2026 materials via the Windows Blog. No independent technical verification is available at this stage. The Windows Blog domain is confirmed live and current.
Aion 1.0 Instruct
is the immediate-term deployment: a smaller, instruction-following SLM Microsoft describes as designed to replace the Phi-4-mini architecture currently handling on-device AI tasks in Edge. It integrates with Windows AI APIs for local inference without cloud dependency.
Unanswered Questions
- What are the minimum RAM and storage requirements for Aion 1.0 Plan's 14B parameters on consumer and enterprise Windows devices?
- Can enterprise IT administrators disable or configure Aion 1.0 Plan through existing endpoint management tools (Intune, Group Policy)?
- What specific Windows AI API integrations are available for Aion 1.0 Instruct at launch, and how does it differ from Phi-4-mini in production?
Aion 1.0 Plan
is the larger model. Microsoft describes it as a 14-billion-parameter local agentic model, intended to ship in-box with Windows, meaning pre-installed, not a separate download. Fourteen billion parameters on-device is a significant hardware commitment. Microsoft hasn’t released minimum system specifications. The specific RAM and storage requirements for the 14B Plan model are unspecified in the announcement; Microsoft hasn’t disclosed them. Teams evaluating deployment across managed Windows fleets should flag hardware specification disclosure as a prerequisite for planning.
The July 2026 open-weights commitment is a scheduled release date, not a completed action. The Instruct variant weights aren’t on Hugging Face yet. Microsoft has committed to making them available, that’s a meaningful signal, but it’s a forward commitment with a one-month-out timeline.
The part nobody mentions in the launch coverage: the on-device model story has two distinct audiences with different risk profiles. Consumer users get a locally running AI that works offline and doesn’t route personal data to the cloud, a privacy benefit. Enterprise IT teams get a 14B model shipping in-box on every new Windows device that may run under managed endpoint policies that weren’t designed with in-box AI workloads in mind. The enterprise governance question for Aion 1.0 Plan is whether IT administrators can disable, configure, or audit it through existing endpoint management infrastructure, Microsoft hasn’t addressed this in the available announcement materials.
What to Watch
Aion 1.0 is the software layer of a coherent Microsoft on-device AI strategy. RTX Spark provides the compute. Aion 1.0 provides the model family. Together they describe a Windows where a meaningful slice of AI inference moves off the cloud and onto the device, with the architectural and governance implications that follow.
Don’t treat the July Hugging Face commitment as a confirmed release until the weights appear. Watch the Windows Insider channel for hardware specification disclosures before fleet deployment planning.