v0.2.13 v0.2.0

high risk

nexu-io/open-designOct 2, 2026 → Oct 2, 20260 days between releases

What's New

Adds Nemotron 3 Diarization, MagpieTTS v2607, and VoiceChat, plus upgrades llama.cpp and changes Vulkan dependencies

Breaking Changes

  • Upgraded llama.cpp: ggml now comes from a single pinned submodule instead of a separate ggml submodule
  • Linux Vulkan archives now use the system C++ runtime, requiring Mesa drivers

New Features

  • Nemotron 3 Diarization support with Sortformer architecture
  • MagpieTTS v2607 support with frame-stacking and Arabic, Korean, and Portuguese
  • Nemotron Labs VoiceChat for full-duplex speech-to-speech
  • Live diarized transcription with endpointing enabled by default
  • Inverse text normalization (ITN) and Text normalization (TN)

Improvements

  • Faster MagpieTTS performance on CUDA and CPU via persistent kernels and fused convolutions
  • Faster CPU GEMMs

Upgrade Notes

Users must update build configurations due to the llama.cpp submodule change. Linux Vulkan users need to install Mesa drivers to support the new system C++ runtime.

The Two Releases

v0.2.13

Sub2API 0.2.13

Oct 2, 2026· 20 days ago

Bug fix release resolving issues with TypeSafe API Key creation and billing settlement.

Full release notes
v0.2.0

v0.2.0

Oct 2, 2026· 19 days ago

Adds support for Nemotron 3 Diarization and MagpieTTS v2607, introduces faster performance optimizations, and includes the new Nemotron Labs VoiceChat for full-duplex speech-to-speech.

Full release notes