NVIDIA Nemotron 3.5 Bolt E4B open on Hugging Face (Gemma-4-E4B post-train, ~4GB GGUF); Build and Ollama day-0

Why it matters

【Key points】

On Oct 23 (Beijing evening) NVIDIA shipped Nemotron 3.5 Bolt E4B foreshadowed in e/887 (Oct 15 Super VL) as “later this week.” HF nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB lists Release Date 10/23/2026, lastModified ~16:00 UTC; @NVIDIAAI posted “now available” ~15:41 UTC linking HF, Build, and Ollama.

Card: post-trained on Gemma-4-E4B for edge; ~4GB Q4_K_M GGUF primary, plus BF16 and DFlash; OpenMDW-1.1, gated=false. Vendor table seats it with Qwen3.5-4B, Gemma-4-E4B, Phi-4-mini. Same day: Build page live; Ollama tag nemotron-3.5-bolt:e4b-nvfp4. No standalone E4B tech post on developer.nvidia.com yet (404); evidence is HF + official X + Build.

【Verifiable】

https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB

https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16

https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-DFlash

https://x.com/NVIDIAAI/status/2113657112408785153

https://build.nvidia.com/nvidia/nemotron-3.5-bolt-e4b

https://ollama.com/library/nemotron-3.5-bolt:e4b-nvfp4

e/887: https://developer.nvidia.com/blog/inside-nvidia-nemotron-3-super-vl/

【Our read】

1. e/887 named a dated “later this week” drop. HF + official X closes it into downloadable weights and runnable entry points.

2. Post-training on Gemma-4-E4B, not a new Nemotron-scale base—compare in the 1–4B edge band. OpenMDW-1.1 + ungated is friendlier than many request-gated edge repos; 4GB GGUF primary points at CPU/local first.

3. Edge assistants, local RAG, and offline inference now have a Nemotron-branded HF + Build + Ollama path next to Super VL. Teams on Gemma-4 mainly review behavior and dual licenses.

4. Limits: no dedicated E4B tech blog; vendor benches; Gemma-4 base means Google + NVIDIA terms both apply. Don’t read “Bolt” as a new architecture generation.