Google · 2026-06-04 · major
Magenta RealTime 2 — Google's Open-Weights Live Music Model Lands With ~200ms Control Latency, Native Apple Silicon C++/MLX Inference, and Both 2.4B and 230M Variants on CC BY 4.0
Google DeepMind's Magenta team open-weights MRT2, a real-time music model with ~200ms control latency (15x faster than v1), MIDI/text/audio steering, and a C++/MLX engine that runs locally on Apple Silicon MacBooks.
An open-weights live music model that responds to MIDI, text, and audio in ~200ms and runs entirely on your MacBook.
Key specs
| Parameters | 2.4B (base) / 230M (small) |
|---|---|
| Control latency | ~200ms |
| Frame size | 40ms |
| Training data | ~71k hours stock music |
| Speedup vs v1 | ~15x |
What is it?
Magenta RealTime 2 (MRT2) is the second generation of Google's open-weights model for continuous, low-latency musical audio generation. It comes in a 2.4B base model and a 230M small model, both released under CC BY 4.0 weights with Apache 2.0 code, plus a plugin bundle you can drop into a DAW. Unlike a typical text-to-music model that emits a finished clip, MRT2 streams music while you steer it live.
How does it work?
Three components: SpectroStream is the audio codec, MusicCoCa is a joint text-audio embedding model, and a decoder-only transformer LLM generates audio tokens. MRT2 switches from v1's 2-second chunks to 40ms frames with frame-level autoregression and sliding-window attention, cutting end-to-end control latency from ~3s to ~200ms. A new C++ inference engine built on Apple's MLX framework runs the base model natively on M3 Pro / M2 Max MacBooks; the 230M model runs on any Apple Silicon, including MacBook Air.
Why does it matter?
MRT2 is the only open-weights model that supports real-time continuous music generation with low-latency control, which makes it usable as an actual instrument rather than a render-and-listen toy. Musicians can play it live via MIDI, prompt it for a style change mid-jam, or fold it into a DAW alongside conventional plugins, all without sending audio to a cloud API.
Who is it for?
musicians, producers, music-software developers, generative audio researchers
Try it
pip install magenta-rt # or grab the MacOS plugin bundle at magenta.withgoogle.com/mrt2