IBM Granite 4.0 1B Speech: #1 on OpenASR Leaderboard at Under 1.5GB VRAM, Apache 2.0 Licensed

IBM / HuggingFace·medium signal

IBM released Granite 4.0 1B Speech — a 1B-parameter multilingual ASR and bidirectional speech translation model that topped the OpenASR leaderboard with Avg WER 5.52 and RTFx 280 real-time factor. It is half the size of its predecessor (granite-speech-3.3-2b), adds Japanese ASR and keyword biasing, and runs on under 1.5GB VRAM. Compatible with transformers ≥ 4.52.1 and vLLM out of the box with Apache 2.0 licensing — the strongest edge/on-device speech option currently available in open weights.