NVIDIA has released the Nemotron 3.5 Bolt E4B 4GB model, an open small language model (SLM) optimized for fast conversational agents, game characters, and robot interfaces that run locally on RTX Spark, GeForce RTX, and Jetson hardware. The launch took place on October 23, 2026, complementing the local AI ecosystem of the RTX Spark platform.
What is Nemotron 3.5 Bolt E4B
The model has about 8 billion total parameters (4B effective), a 64K context window, and was post-trained from Google’s Gemma 4 E4B using reinforcement learning only. It is designed to fit in roughly 4 GB of VRAM, leaving room for speech and vision models on the same device. In NVIDIA tests with Q4_K_M quantization via llama.cpp, it uses between 3.1 and 4.3 GB on RTX cards.
The company highlights superior performance in conversational benchmarks compared to similar-sized models such as Gemma 4 E4B and Qwen3.5-4B. It can be tested on platforms like SkyrimNET, an AI NPC mod for The Elder Scrolls V: Skyrim, starting October 23.
Context in the RTX Spark ecosystem
The Bolt E4B arrives alongside RTX Remix updates (including DLSS 5 support) and the Personal AI Router (PAIR) now generally available. The RTX Spark platform, with up to 128 GB of unified memory and up to 1 petaflop of AI performance, is positioned by NVIDIA as a compact Windows PC for creation, AAA gaming, and local agents simultaneously.
Developers can use the model for more responsive NPCs, offline voice assistants, and robot interfaces without relying exclusively on the cloud. The model is available in open format, compatible with tools such as llama.cpp, Ollama, and others.
This release reinforces NVIDIA’s strategy of bringing AI agents to the edge and personal PCs, especially useful for mod creators and game developers seeking low-latency local interactivity.
Image credit: NVIDIA
By GeekikiBot
