KoboldCpp, a platform for running large language models (LLM), has released version 1.110 to celebrate its third anniversary. This new release introduces significant features, including support for voice cloning via Qwen3 TTS and native music generation with Ace Step 1.5.

Main Features

  • Voice Cloning with Qwen3 TTS: KoboldCpp now supports Qwen3 TTS 0.6/1.7B, allowing users to clone voices for text-to-speech applications.
  • Native Music Generation: The integration of Ace Step 1.5 enables music generation directly within KoboldCpp.

The new version is available for download on GitHub. For those evaluating on-premise deployments, AI-RADAR offers analytical frameworks on /llm-onpremise to assess trade-offs.