Canary-Qwen-2.5B

Canary-Qwen-2.5B is NVIDIA's NeMo hybrid model that fuses automatic speech recognition with a large language model. A FastConformer speech encoder feeds audio tokens through adapters into a Qwen3-1.7B decoder, so the same model transcribes English speech with punctuation and capitalization and then, in LLM mode, can summarize the transcript or answer questions about it.\n\nAt release it topped the Hugging Face OpenASR leaderboard with a record 5.63% word error rate while running at 418x real-time on just 2.5 billion parameters, trained on 234,000 hours of diverse English speech data. It is licensed CC-BY-4.0, ready for commercial use.\n\nThe hybrid design collapses the traditional transcribe-then-process pipeline into one deployment, useful for meeting transcription, call analytics and voice-assistant front ends.