Now supported in all major browsers (2026)

WebGPU

The modern GPU API for the web

High-performance graphics and compute. The foundation for the next generation of browser-based AI, games, and scientific computing.

Explore AI Use Cases View Live Demos

What is WebGPU?

WebGPU is the next-generation graphics and compute API for the web. It gives web developers low-level, high-performance access to the GPU — similar to Vulkan, Metal, and Direct3D 12 on native platforms.

Unlike its predecessor WebGL (based on OpenGL ES), WebGPU was designed from the ground up for modern GPUs. It offers:

  • First-class support for general-purpose GPU compute (GPGPU)
  • Better performance and lower overhead
  • Modern features: bindless resources, compute shaders, advanced memory management
  • A new shading language: WGSL (WebGPU Shading Language)
🚀
Successor to WebGL
2023–2026 rollout
Graphics Excellent
Compute / ML Best-in-class
Browser Support ~95%+ (2026)

A Brief History

From early experiments in 2016 to universal browser support in 2026

JUNE 2016
Google presents “Explicit web graphics API”

First public discussion about replacing WebGL with a modern, low-level API at the WebGL working group.

FEBRUARY 2017
W3C “GPU for the Web” Community Group formed

Apple proposes WebGPU (initially inspired by Metal). Google, Mozilla, and Apple begin collaboration. Khronos hosts “WebGL Next” meeting.

2020–2021
WGSL chosen & first spec publication

WebGPU Shading Language (WGSL) standardized. First public Working Draft published.

APRIL 2023
Chrome 113 & Edge ship WebGPU

First stable browser implementations. Major milestone for real-world adoption.

2025
Safari 26 & Firefox 141 ship WebGPU

Apple and Mozilla bring full support. WebGPU reaches critical mass.

EARLY 2026
Universal support across all major browsers

W3C Candidate Recommendation. WebGPU becomes production-ready for graphics and AI workloads.

WebGPU in AI & Machine Learning

WebGPU’s real power for most developers lies in its compute capabilities. It enables high-performance machine learning inference directly in the browser using the user’s GPU.

Why WebGPU for AI?

  • Up to 50–100× faster than WebAssembly/CPU
  • Runs quantized LLMs, vision models, embeddings locally
  • No server round-trips = lower latency & cost
  • Full privacy — data never leaves the device
🔒

Self-Hosted AI in the Browser

True client-side inference

Run powerful AI models entirely on the user’s device using WebGPU. Perfect for privacy-sensitive applications, offline tools, and reducing cloud costs.

Examples
• Local LLM chatbots (Llama 3, Mistral)
• Real-time image generation
• On-device speech-to-text
• Private embeddings & RAG
Key Benefits
✓ Zero API costs
✓ Works offline
✓ No data exfiltration
✓ Instant inference

Featured Projects & Demos

Production-ready libraries and demos using WebGPU for AI

WebLLM

MLC AI

High-performance in-browser LLM inference engine. Runs Llama, Mistral, Phi and more entirely in the browser with WebGPU acceleration. OpenAI-compatible API.

WebGPU Quantized models

Transformers.js

Hugging Face

Run thousands of Hugging Face models in the browser. Native WebGPU support via ONNX Runtime. Up to 100× faster than WASM for many tasks.

Text, Vision, Audio 100+ models

Official WebGPU Samples

Official collection of interactive demos showcasing graphics, compute shaders, and advanced techniques. Excellent starting point for learning the API.

Browse All Samples →
Also explore: ONNX Runtime WebTensorFlow.js (WebGPU backend)
Key Resources
WebGPU is a W3C Candidate Recommendation (2026).
Built with ❤️ for the open web.