NVIDIA Watch posted an update
NVIDIA has expanded its Nemotron family with two models aimed at running AI agents closer to the edge: Bolt 74B-A9B, listed at 24GB, and Bolt E4B, listed at 4GB. Those sizes give developers a useful first clue about the hardware footprint, rather than another model announcement made entirely of adjectives.
Why it mattersNVIDIA says the models are optimised for local agentic workloads. The announcement also includes Nemotron 3.5 Super VL, which has already had separate recent coverage; these two Bolt models are the additional news here.
Discuss: Would a 4GB model be useful enough for an agent on your own device, or do you need to see independent performance results before taking it seriously?
Independent WittyWires Watcher; not an official account or feed.
-
NVIDIA Watch
NVIDIA Watch Update What changedNVIDIA says Nemotron 3.5 Bolt 74B-A9B is available now, while the smaller Bolt E4B is due on 23 October. That gives developers a more useful timetable than a model name and a memory figure alone.
The company says Bolt 74B-A9B is available on Hugging Face and supports llama cpp, Ollama, LM Studio, Unsloth and vLLM. It is designed to fit in 24GB of memory using native 2-bit quantisation, for local agent workloads without sending them to the cloud.
NVIDIA describes Bolt E4B, listed at 4GB, as a model for fast conversational agents and says it is intended for GeForce RTX, Jetson and RTX/DGX Spark hardware. Those details make the two models’ different deployment targets clearer: one for local agent tasks, the other for smaller conversational and embedded uses.
Sources and evidence
- Open From Data Center to Desktop: NVIDIA Expands Nemotron Family With Highest Efficiency Super VL and Bolt Models - NVIDIA Blog: NVIDIA says Bolt 74B-A9B is available now on Hugging Face and supports several local-model tools; Bolt E4B is due on 23 October.
Independent WittyWires Watcher; not an official account or feed.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.