Model components / Inference code

Apple Core ML Runtime

On-device inference runtime that executes models across Apple's heterogeneous compute (CPU, GPU, and Neural Engine) on iPhone, iPad, Mac, and Vision Pro. The compiler automatically partitions model graphs across available hardware for optimal latency and power efficiency. Powers Siri, on-device autocorrect, image recognition, and Apple Intelligence features. The most widely deployed on-device inference runtime by install base (2B+ active Apple devices).

Apple Core ML on-device inference runtime (iOS/macOS/watchOS/tvOS, Neural Engine + GPU + CPU). Proprietary system framework; companion coremltools (conversion) is open (BSD-3-Clause). Confirmed live June 2026 via Apple Developer docs.

Openness

1 high confidence

1.0

runtime: proprietary(Apple system framework, closed)
conversion-tooling: coremltools(BSD-3-Clause, OSI, open)
model-format: .mlmodel/.mlpackage(documented)
hardware: Apple-silicon-only(Neural Engine)

The inference runtime itself is a closed Apple framework; only coremltools (model conversion) is open BSD-3. Headline (the runtime) is closed; scored closed with open-tooling note.

https://developer.apple.com/documentation/coreml recorded 2026-06-04
Core ML is Apple's proprietary on-device ML inference framework; runtime closed, runs models on Apple hardware

Adoption

5 medium confidence

5.0

Core ML is the system on-device inference runtime shipped on every modern iPhone/iPad/Mac (>1B active Apple devices) and used by a large share of App Store ML features. Mass-market reach attributed to the Apple device install base the runtime ships on, not a developer download count.

https://developer.apple.com/documentation/coreml recorded 2026-06-04
Core ML runs on-device across iOS/macOS/watchOS/tvOS, the default on-device inference framework for Apple platforms

Capability

4 medium confidence

4.0

Best-in-class on-device/edge inference runtime (Neural Engine acceleration, quantization, broad converted-model coverage) but constrained to Apple silicon and on-device scale; C4 within this category (frontier C5 reserved for the large-scale server engines vLLM/SGLang/TensorRT-LLM).

https://developer.apple.com/documentation/coreml recorded 2026-06-04
Neural Engine acceleration, on-device inference, model optimization features

Unchanged since 2026-06-09 (last edited, not re-checked)