v0.6
CaviraOSS/LongMemoryv0.6Aug 18, 2025by liPatrick
AI Summary
Ultravox v0.6 introduces new gemma3 and qwen3 model variants, alongside significant improvements in Hindi language understanding and audio robustness. The release also adds support for VoiceBench evaluation and allows for response generation without a user message.
Key Highlights
- New model variants: gemma3 and qwen3
- Improved Hindi language understanding
- Enhanced audio robustness and noise handling
- New eval support for VoiceBench
New Features
- New gemma3 and qwen3 variants
- VoiceBench evaluation suite integration
- Response generation with no user message
Full Release Notes
We're releasing **Ultravox v0.6** today. The [weights](https://huggingface.co/fixie-ai) have been pushed to Hugging Face. If you're using the [Ultravox Realtime APIs](https://docs.ultravox.ai), v0.6 is the new default. ## What's New * v0.6 improves upon 0.5 in the following ways: * Improvements on Hindi language understanding. * Produces <noise> token on noisy or non-human audio. * Improved background noise and audio quality robustness in all languages. * New gemma3 and qwen3 variants in addition to our base llama 3.3 models. ## Evals New eval support for [VoiceBench](https://github.com/MatthewCYM/VoiceBench). The benchmark tests speech-language models on 9 different tasks ranging from open-form text generation to question-answering and instruction following. ## Training This version of Ultravox continues to use a frozen Llama pre-trained core (3.1 for 8B and 3.3 for 70B), along with new gemma3 27b and qwen3 32b variants. ## What's Changed Training Stability: Patch HF Hub and Datasets methods and update [datasets.py](http://datasets.py/) by @farzadab #280 General improvement. by @zqhuang211 #281 Ultravox v0.6 + general improvements by @liPatrick #309 Allow response generation with no user message by @matthewclso #310 Add voicebench evaluation suite by @zqhuang211 #312 **Full Changelog**: https://github.com/fixie-ai/ultravox/compare/v0.5...v0.6