Nemotron 3 Ultra

Nemotron 3 Ultra is NVIDIA's orchestration-focused model in Alfrada. Pick it when you want a strong Workhorse model for long-running planning, multi-tool execution, and sessions where the model needs to keep track of a moving plan without jumping to SOTA-band pricing.

What It Is

• Picker name: Nemotron 3 Ultra [Global]
• Provider route: OpenRouter global route
• Model family: NVIDIA Nemotron 3
• Architecture: 550B total / 55B active hybrid Transformer-Mamba MoE
• Context window: 256K tokens in Alfrada
• Tier: Workhorse
• Modes: Agent, Auto, and Swarm

Use Nemotron 3 Super 120B [EU] instead when EU-hosted routing is the main requirement. Use Nemotron 3 Ultra when global routing is acceptable and you want the larger model for harder orchestration work.

Best Fit

Nemotron 3 Ultra is strongest when the job has several steps and the value comes from keeping the work organized:

• planning a project before execution
• coordinating research, analysis, and artifact creation
• running tool-heavy workflows with many intermediate files
• leading or participating in Swarm sessions
• producing structured briefs, launch plans, and technical work plans

It is not the default pick for every chat. If you only need a quick rewrite, summary, or small lookup, a faster model is usually a better fit.

When To Pick It

Pick Nemotron 3 Ultra when you want:

• planning depth without SOTA cost - a capable model for longer execution loops at Workhorse pricing
• tool confidence - a model comfortable with search, files, code, documents, presentations, and multi-step outputs
• Swarm orchestration - a strong lead model or specialist worker for decomposed work
• large working memory - enough context for long briefs, source sets, and iterative artifact chains

If the task is legally sensitive, extremely high-stakes, or demands the absolute strongest reasoning available, compare it with the current SOTA picks before committing the whole session to one model.

Good Prompt Patterns

Long-running project plan

Use Nemotron 3 Ultra. Build a step-by-step launch plan for [product]. Research the market, map the buyer, identify risks, then produce a memo and a slide outline. Keep the plan visible as you work.

Nemotron is a good fit when the session needs planning, research, synthesis, and multiple deliverables.

Tool-heavy analysis

Switch to Nemotron 3 Ultra and analyze these uploaded files. Extract the important facts, check anything missing with web research, then create a concise executive brief with clear next actions.

The model has enough context and tool discipline for large source sets and artifact handoffs.

Swarm lead

Run this as a Swarm with Nemotron 3 Ultra leading. Have workers gather evidence, critique assumptions, and return a board-ready synthesis with sources and open questions.

Nemotron works well as an orchestrator when a job benefits from parallel research and review.

Practical Guidance

Start with a concrete outcome, not just a topic. Nemotron performs best when you tell it what the finished work should look like: a memo, deck outline, research brief, implementation plan, or decision note.

Give it constraints early:

• audience
• depth
• sources or files to use
• time horizon
• output format
• what to avoid

For example: "Build a 900-word investment memo for a non-technical founder, cite sources, and separate facts from assumptions" is better than "research this company."

Heads Up

• Global route: Nemotron 3 Ultra [Global] uses OpenRouter's global path. Use Nemotron 3 Super 120B [EU] when EU residency matters.
• Workhorse tier: It is priced and ranked as a Workhorse model in Alfrada, not a SOTA model.
• Auto can select it: It is part of the Auto routing pool, so you may get it automatically when Alfrada sees a planning-heavy job.
• Swarm eligible: You can use it in Swarm for orchestration or specialist-worker roles.

For the launch announcement, see Nemotron 3 Ultra Is Live.