Nemotron 3.5 Bolt E4B 4GB
NVIDIA released Nemotron 3.5 Bolt E4B 4GB, an open small language model for local voice agents, game characters and robots that runs in about 4GB of VRAM.
It is post-trained from Google's Gemma 4 E4B using reinforcement learning only, with no supervised fine-tuning, because NVIDIA says SFT helped tool calling but hurt conversational quality and verbosity. It adds multi-turn tool use, a conversational_mode flag for short speech-ready replies, and configurable reasoning with a runtime thinking budget. NVIDIA calls it the third-generation small model for NVIDIA ACE game characters.
- Date
- Friday 23 October 2026
- Lab
- NVIDIA
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Parameters | 8B total, 4B effective | company |
| Context window | 64K tokens | company |
| VRAM, Q4_K_M build on RTX | 3.1 to 4.3 GB NVIDIA tests | company |
Based on MarkTechPost coverage of the launch post; the license was not stated in the text read.
Sources
This record was partly confirmed: some claims could not be checked on 24 October 2026. How we check