k
keynews.ai
keynews.ai
FrontierThe story, in brief

Kyutai Releases MuScriptor: An Open-Weight Decoder-Only Transformer for Multi-Instrument Music Transcription to MIDI

Kyutai's new open-weight model does what closed systems couldn't: turn any song into playable MIDI across all instruments.

Reporting by MarkTechPost 1 min brief
Illustration of a transparent lens revealing connected networks across layers of paper.
Exploring the next frontier of AI research.AI illustration by KeyNews
The KeyNews take

Why it matters

A capable open-source music transcription model trained on 1.45M synthetic examples demonstrates how synthetic data scales transformer capabilities beyond audio-to-text tasks—directly challenging closed commercial offerings and expanding the frontier of what open weights can do.

The key facts

10 to know

  1. MuScriptor: decoder-only Transformer architecture from Kyutai and Mirelo

  2. Trained on 170k real recordings + 1.45M synthetic MIDI examples

  3. Multi-instrument MIDI transcription from full mixes

  4. Open-weight release (vs. closed YourMT3+ baseline)

  5. Three-stage pipeline with instrument conditioning

  6. Benchmarked against YourMT3+

  7. Open-weight decoder-only Transformer architecture

  8. Trained on 170k real recordings + 1.45M synthetic MIDIs

  9. Multi-instrument transcription to MIDI capability

  10. Interactive demo available

Go to the source

MarkTechPostmarktechpost.com

Publisher excerpt: MuScriptor is an open-weight, decoder-only Transformer from Kyutai and Mirelo. Trained on 170k real recordings plus 1.45M synthetic MIDIs, it transcribes full multi-instrument mixes into MIDI. The article covers its three-stage pipeline, benchmarks against YourMT3+, instrument conditioning, and…
Read original report
Back to today's editionMore frontier news

The wider picture

View all
Illustration of a transparent lens revealing connected networks across layers of paper.
AI illustration by KeyNews
Frontier01

Alibaba Unveils Zhenwu V900 — and Plans Qwen Models With Up to 10 Trillion Parameters

Alibaba is advancing on two fronts simultaneously: announcing a custom AI accelerator (Zhenwu V900) and committing to massive model scale (10T parameters for future Qwen releases). For practitioners, this matters as a credible third-party capability play outside the US-China licensing squeeze; for enthusiasts, it's a significant lab-race signal about training compute and parameter scaling as competitive levers.

TechRepublic
Illustration of a transparent lens revealing connected networks across layers of paper.
AI illustration by KeyNews
Frontier02

Kyutai Releases Voice of Reason: A Speech-Native Model that Solves Spoken Math with Reinforcement Learning

A new capability frontier: models that reason directly in speech without transcription bottlenecks. This changes how we think about multimodal reasoning and what's possible with open-weight releases at scale.

MarkTechPost
Illustration of a transparent lens revealing connected networks across layers of paper.
AI illustration by KeyNews
Frontier03

Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war

Frontier labs are shipping upgraded reasoning and multimodal models in rapid succession, signaling acceleration in the capability race. Simultaneous price cuts reshape AI economics for practitioners.

Simon Willison