Nemotron Lightning Model
I played around with the Nemotron Lightning 30b model this week. It is one of the most affordable models on the market, at approximately 25% of the price of OpenAI’s ChatGPT Luna model. At the time of this article, Nemotron Lightning costs just $0.05 per million input tokens and $0.20 per million output tokens.
I tested it for my usecase of proofreading my writings. And, well, it did terribly. I had a coding agent evaluate it against DeepSeek v4p1 Flash (also ~4x the price) as a baseline, and what my AI said about Nemotron
The judge found DeepSeek’s responses clean and actionable. Nemotron correctly caught several issues, but repeatedly added non-errors, style changes, invalid “no change” rows, and one hallucinated Markdown-link correction.
and
The repeated runs already explain the odd result: Nemotron is not merely slower. It alternated between a 610-token answer and an eight-token “no errors” answer for the identical request, so its average latency is dominated by unstable output behavior. DeepSeek has been consistent; I’m completing the GLM runs now.
and
Nemotron’s two long answers took 1.306s and 2.381s. Its eight short responses averaged 0.805s, although they were incorrect because the file clearly contains errors. That explains the previous “slow” run: it generated a very long, low-quality table, while DeepSeek generated about 50 tokens of useful corrections.
However, I did find a use case for it: An AI dictionary inside of Emacs. I have a Emacs function named jc/whatis and I type in a word I want a definition for and I get back the word definition almost instantly.
Nvidia says the Nemotron Lightning is for “designing AI Agent systems, chatbots, RAG systems, and other AI-powered applications.” This is so vague as to be useless. I guess due to its low cost for cached tokens, it might be good for ad-hoc conversations. I haven’t built such a chatbot yet, but from what I’ve seen I suspect it is not going to be worth it.