NVIDIA's new 4GB Nemotron model is built to run game characters and voice assistants on your own machine.

On Friday, Oct 23, NVIDIA released Nemotron 3.5 Bolt E4B 4GB, a small open model "built for local assistants, interactive characters and robots that can carry on a conversation and use tools" (NVIDIA AI on X).

According to its model card, NVIDIA built it on top of Google's Gemma 4 E4B, and it comes already compressed to a 4-bit format so the file is about 4GB (Hugging Face). It takes text in and gives text out, with up to 64K tokens of context. You can turn its step-by-step thinking on or off, cap how long it thinks, or use a conversational mode for chat. NVIDIA reports 55.60 on MultiChallenge, a test of holding a multi-turn conversation, against 39.36 for the Gemma model it started from, and 71.21 against 60.51 on RoleBench, a role-play test.

NVIDIA's prompting guide even includes examples of setting it up as a game-style character. The license is NVIDIA's OpenMDW-1.1 (with Gemma 4 under Apache 2.0), and NVIDIA says it is "ready for commercial or non-commercial use." It is on Hugging Face and build.nvidia.com, and Ollama and Unsloth have both posted versions to run locally (Ollama on X, Unsloth on X). Unsloth says it runs in about 6GB of RAM.

Try it: if you make games or interactive projects, load it locally and give it a character sheet. A model this small can sit next to a game on an ordinary gaming PC, so you can test NPC dialogue without paying per message.