Nemotron 3.5 Lightning: NVIDIA's Newest Model, Free on BrahmAI

NVIDIA's Nemotron family has been on BrahmAI since July. The newest member, Nemotron 3.5 Lightning, is now live — and like Nano and Super before it, it is available on every plan, including the free one.

No card, no trial period.

Where it sits in the family

Nemotron 3 came in three sizes. 3.5 Lightning is a new generation rather than a fourth size:

Model Best for On the free tier
Nemotron Nano Fast, cheap, everyday questions
Nemotron Super Balanced — 120B mixture-of-experts
Nemotron 3.5 Lightning Newest generation, fast, reasoning
Nemotron Ultra 550B flagship, hardest problems Expert / Pro

Lightning is not a replacement for Ultra. Ultra remains the family's flagship and stays on the paid tiers. Lightning is the newest architecture — quicker than Ultra, and strong enough that it earns a place next to Nano and Super rather than below them.

It reasons by default

Like the rest of the family, Lightning thinks before it answers rather than responding in one pass. On a simple test question it spent around 210 tokens reasoning before writing anything.

That is usually what you want for anything involving logic, code, or multi-step questions. If you would rather have a fast one-pass answer, the Thinking control in the model picker turns it off.

A long memory

Lightning holds a large amount of context — enough for long conversations, or a sizeable document, without losing the thread of what came earlier.

That matters more than raw benchmark scores for everyday use. A model that forgets the start of your conversation is frustrating regardless of how clever it is.

No catch on the free tier

Worth being straightforward, because "free AI" usually has one.

Lightning on BrahmAI's free plan is the real model, not a trimmed-down variant or a trial. There is no message cap that suddenly bites, no quality downgrade after a few turns, and no card required.

Nemotron Ultra is the one member of the family that stays on the paid plans — it is the 550B flagship, and it belongs there. We would rather be clear about that than put every model on the free tier and quietly ration them.

What it is good at, and what it is not

Good at: everyday questions, coding help, reasoning tasks, long documents where the context window matters.

Not for: images. Every Nemotron model is text-only. Attach a photo and the model will not see it — use one of the vision-capable models instead, of which BrahmAI has 28. PDFs and text documents work fine, because those are converted to text before the model sees them.

Trying it

Nemotron 3.5 Lightning is in the model picker under NVIDIA, marked Fast. It works in Chat, in the Council, and as an AI Search summariser.

If you are on the free plan it is already available — nothing to unlock. Paid plans get it too.

The wider point

Free tiers on AI products are usually a demo: a handful of messages, an old model, or a queue.

Ours runs on real, current models from named companies — NVIDIA, Google, DeepSeek, Alibaba. Not last year's release, and not a cut-down variant.

If a model ever stops making sense on the free tier, we move it to a paid plan rather than degrade it quietly and hope nobody notices.