Bijan Bowen tests Nemotron 3.5 Lightning, a 30-billion-parameter mixture-of-experts modelA mixture-of-experts model contains multiple specialized subnetworks and activates a selected subset for each input instead of using every parameter every time. with roughly three billion active parameters, a Mamba 2 architecture and a one-million-token context windowA context window is the maximum amount of tokenized information an AI model can consider during one processing session.. The model runs quickly on local hardwareOn-device inference runs a trained model on the user's device rather than relying entirely on a remote model service. and is designed to use speculative decodingSpeculative decoding accelerates text generation by proposing several tokens cheaply and checking them together with the target model. for higher throughput.
In coding and browser-driven tasks, Bijan Bowen finds uneven results. Nemotron 3.5 Lightning can install dependencies, call tools and keep working through an agent loop, but its games, visual interfaces and browser interactions often need intervention or remain incomplete.
Long-context retrievalRetrieval is the process of selecting relevant stored information and returning it to an AI system for the current task. is the clearer strength. Bijan Bowen gives the model a document of roughly 60,000 tokens and sees it recover a requested detail accurately, supporting its use for large repositories, document sets and agent memory where recall matters more than polished front-end output.
Bijan Bowen concludes that Nemotron 3.5 Lightning is not a coding powerhouse, but its speed, tool use, personality and large context window make it a useful local workhorse for agentic tasks that can tolerate supervision.
Watch on YouTube




