Nvidia has released Nemotron-Labs-Audex-30B-A3B, a unified audio-text large language model built on the Nemotron-Cascade-2-30B-A3B text-only MoE backbone. The model extends the original architecture with an audio encoder for inputs and discrete audio tokens for outputs, enabling capabilities in speech recognition, translation, text-to-speech, and audio generation.
- It retains the reasoning, alignment, knowledge, long-context, and agentic capabilities of its text-only predecessor with marginal or no regression.
- The model supports a 1M-token context length and operates in both "thinking" and "instruct" (non-thinking) modes.
- A smaller Nemotron-Labs-Audex-2B variant is also available.
This release provides a unified solution for multimodal audio-text tasks while preserving the strong performance characteristics of the underlying Cascade model.