Metal is the current generation of low-level graphical API that Apple uses, which powers the hardware accelerated graphics on their OSes, namely macOS. Metal 1.0 was introduced on September 30th, 2015, alongside Mac OS X El Capitan and iOS 8.

Before the Apple Silicon transition, the Metal API received regular major updates every 2 years. The major chain of updates now come around once every 3 years instead. The very first version of Metal required at least an Apple A7 chip or better. It also required at least a 2012 Mac GPU or better.

As of June 2025, Metal 4 was announced alongside macOS Tahoe and iOS 26, and is going to be a major update. This API is built from the ground-up for Apple Silicon only, leaving behind Intel Macs altogether. This was mentioned in the “Platform State of the Union”, so it appears both the newer and older API will ship with macOS Tahoe as macOS 26.0 retains support for some Intel Macs.

• See: Understanding the Metal 4 core API – Apple

(Above: A frame from the WWDC ’25 Metal 4 API video, appearing to be a rendering.)

Requirements

• Metal 4 requires an Apple M1 or an A14 Bionic or later.
  .

Key points about Metal 4

Many improvements over the same technologies introduced in Metal 3 such as hardware accelerated ray tracing, etc; which are listed below.

• OS Implementation: macOS, iOS, iPadOS, tvOS, visionOS, and watchOS
  .
• Adds MetalFX Frame Interpolation – generates an intermediate frame for every two input frames to achieve higher frame rates or a more stable one. It is a similar technique to DLSS and FSR frame generation.
  .
• Adds MetalFX denoising – real-time ray-tracing and path tracing. Helps achieve the best trade-off between quality and performance with fewer cast rays. “The new MetalFX API enhances real-time ray-tracing rendering pipelines by integrating denoising directly into the upscaling process..” as mentioned on Apple’s website.
  .
• Upscale Rendering – machine-learning based upscaler. “New this year, the MetalFX temporal upscaler supports dynamically sized inputs, so you can dynamically lower the input resolution for particularly complex frames” as mentioned on Apple’s website.
  .
• Introduces MTLTensor: “a multi-dimensional data structure designed to facilitate sophisticated machine learning workloads on the GPU”. See: app.daily.dev
  .
• The metal shading language is what controls GPU execution. It is C++ based.
  .

Some recent related news

• See: Apple Silicon machine learning code may become more easily portable to Nvidia hardware – AppleInsider – 7/15/2025

According to LaurieWired on 7/15/2025: “MLX, Apple’s machine learning framework, just merged a CUDA Backend. If you prototype models on a Mac, you can now ship the same code to (Linux) NVIDIA boxes. This isn’t some sort of JIT or translation layer, it’s natively executing on official CUDA runtimes.”

This doesn’t at all mean you can just start plugging in Nvidia GPUs to your Macs, unfortunately. Developers can now write and test code locally before putting it onto CUDA-based Nvidia hardware. This is still a work in progress having started March 21st, is not very usable yet, but is an Apple-sponsored project.
.

In Conclusion

Apple is pushing hard in the compute space, and also in gaming for the Mac. This is just a first glance at a technology which is evolving, as noted by the recently related news. We want our readers to understand the new technologies involved in the new Apple OSes, especially in the Mac and in the iDevices. This is just a summarization of available information, and could be subject to change.