Skip to content

Benchmark

Local AI Hardware Requirements

Summary

use.fo Local AI / Local STT requires macOS Apple Silicon (M1 or later) or Windows x86/x64 with VRAM greater than 4GB. macOS Intel, iOS, Android, and Windows ARM are not supported. Performance scales with available VRAM and neural engine capability.

Reproducibility note

This is a first-party benchmark. Use the methodology, device conditions, model details, aggregate result, and limitations below to understand or challenge the decision. It is directional evidence, not an independent certification.

What we tested

Published Local AI compatibility boundaries, plus the disclosed MacBook Pro M2 Pro / 64 GB reference setup used for the first-party Auto Research STT-to-answer evaluation.

Limitations

Note

No Windows device, GPU tier, macOS version, model-load time, or VRAM-use trace is included in the disclosed benchmark record.

Note

Compatibility does not establish a particular speed or quality level. Performance can vary with hardware, thermal state, system load, model build, speaker, and audio environment.

Note

macOS Intel is not supported. If you have an Intel Mac, use Subscription or BYOK.

Note

Windows ARM is not a current Local AI target. ARM devices can use Subscription and BYOK.

How to choose

Choose Local AI only if you have an Apple Silicon Mac or a Windows x86/x64 device with more than 4 GB VRAM and prefer on-device processing. The published M2 Pro result explains the selected Apple Silicon route, but it is not a speed promise for other devices; test your own workflow before purchasing. For iOS, Android, Intel Mac, or Windows ARM, use Subscription or BYOK.

Methodology

  • Reference setup: MacBook Pro with Apple M2 Pro and 64 GB memory. This is the only disclosed benchmark device in the current record.
  • Compatibility boundaries are product support facts, not benchmark scores: Local AI supports Apple Silicon Mac and Windows x86/x64 with more than 4 GB VRAM.
  • The local STT methodology page reports a first-party Apple Silicon selection result for its chosen MLX Audio Whisper Large V3 Turbo configuration. It does not establish a cross-chip, Windows, or device-tier speed ranking.

Results

MetricResultNotes
Disclosed reference deviceMacBook Pro — M2 Pro, 64 GB memoryUsed for the first-party Auto Research STT-to-answer evaluation.
Disclosed local selection timing1,870.97 ms average STT timeSelected MLX Audio Whisper Large V3 Turbo route on the disclosed first-party Apple Silicon benchmark; not a prediction for every Mac.
macOS Apple SiliconSupportedProduct compatibility boundary; no cross-chip speed ranking is published here.
macOS IntelNot supportedInsufficient performance characteristics for on-device inference at this time.
Windows x86/x64, more than 4 GB VRAMSupportedProduct compatibility boundary; no device-specific speed figure is published here.
Windows ARMNot a current targetWindows ARM is not supported for Local AI at this time.
iOS / AndroidNot supportedMobile hardware does not meet minimum requirements for on-device inference.

FAQ

Your Windows laptop qualifies if it has an x86/x64 CPU and a GPU (discrete or integrated) with more than 4GB of VRAM or shared GPU memory. Check Task Manager → Performance → GPU for your VRAM. If it shows less than 4GB, Local AI may not load.

Apple Silicon Mac is the supported macOS boundary. This page does not publish a comparative M1, M2, or M3 performance score; test the app on your own workflow if response time is important.

Local AI requires a Windows x86/x64 device with more than 4 GB VRAM. Check your actual GPU and available VRAM before buying; this page does not publish an integrated-GPU performance benchmark.

No. Windows ARM is not a current Local AI target. You can use Subscription or BYOK on a Surface Pro X.

Evidence and updates

Facts checked on July 24, 2026

Reviewed by use.fo product team. Sources below identify whether a fact comes from an official product page, a first-party test, customer evidence, or an independent source.

  • use.fo Product Facts — plans, platform matrix, providers, data routes, unsupported features; checked July 24, 2026 (official).
  • use.fo Pricing — license price, license scope, provider cost disclosure; checked July 24, 2026 (official).
  • Local AI hardware requirements — Windows architecture, VRAM requirement, Apple Silicon support; checked July 24, 2026 (first party test).
Update log: July 24, 2026 — Published the no-subscription product fact registry and clarified BYOK, Local AI, and Windows support boundaries.