Confirmed
NVIDIA releases Nemotron-3.5-Bolt-E4B, a small model post-trained from Gemma-4-E4B
The 4B-class model targets low-memory voice and device assistants with improved multi-turn tool use and instruction following. It is available in GGUF and BF16 versions, with a companion DFlash drafter for speculative decoding, and is licensed for commercial and non-commercial use.
Every outlet that ran it 1 article
r/LocalLLaMACommunityLead