NVIDIA ships Nemotron 3.5 Bolt E4B, an open vision model that runs on 6GB RAM
danielhanchen · x · 2026-10-23
NVIDIA released Nemotron 3.5 Bolt E4B, an open 8B (4B effective) vision-capable model that runs locally on 6GB RAM devices including phones. Post-trained on Gemma 4 E4B with boosts for agentic use and coding, it targets local assistants, interactive characters, and robots.
- Tops the MultiChallenge multi-turn benchmark at 56.03%, 12.8 points clear of Gemma 4 E4B; NVIDIA's chart shows 70.2 composite with 250ms time-to-first-token.
- Demoed driving a Reachy Mini robot (Bolt for conversation, Cosmos for vision) on Jetson Orin Nano 8GB.
- Weights, data, and recipes are open; Unsloth shipped day-zero GGUF quantizations and a guide (4-bit on 6GB, full precision on 16GB).
More from Infra
- PsiQuantum's 12-inch thin-film BTO wafer spotted in person, and it looks stunning — clarejtbirch · 2026-10-24
- BofA's data center map: servers cost $31.4M per MW, dwarfing every other component — TiernanRayTech · 2026-10-24
- Chip-Backed Debt Is on Track to Become a Trillion-Dollar Market, Concentrating Risk — dinabass · 2026-10-24
- Applied Compute Featured on NVIDIA GTC Keynote for Enterprise Post-Training Platform — rhythmrg · 2026-10-24
- OpenAI details how it scaled ChatGPT storage with multi-tenancy and global caching for 12x faster reads — OpenAI News · 2026-10-24
- Google Cloud's API Gateway now supports streaming endpoints for LLM routing — rseroter · 2026-10-24