Fireworks AI model
NVIDIA Nemotron 3 Ultra NVFP4
Nemotron-3-Ultra-550B-A55B-NVFP4 is a frontier-scale large language model (LLM) trained by NVIDIA, designed to deliver strong agentic, reasoning, and conversational capabilities. It is optimized for the most demanding workloads, including complex multi-step agents, long-context analysis, and high-accuracy reasoning over code, math, and science. The model employs a hybrid Latent Mixture-of-Experts (LatentMoE) architecture, utilizing interleaved Mamba-2 and MoE layers, along with select Attention layers. Like the Super model, the Ultra model incorporates Multi-Token Prediction (MTP) layers for faster text generation and improved quality, and it is trained using an NVFP4 pre-training recipe to maximize compute efficiency. The model has 55B active parameters and 550B parameters in total.
- Provider
- Fireworks AI
- Availability
- Available
- Model ID
- nemotron-3-ultra-nvfp4
- Snapshot
- 2026-10-07
NVIDIA Nemotron 3 Ultra NVFP4 Benchmarks & Agentic Accuracy
Verified primary research & developer preference metrics
What This Means in Real Software Development
- Optimized for NVIDIA Blackwell hardware with cutting-edge NVFP4 4-bit floating point precision.
- 58.2% on SWE-bench Verified, excelling at systems programming, CUDA kernels, and C++ refactors.
- 77.0% on AIME confirms strong mathematical and logic reasoning.
- Available on Fireworks AI with hardware-accelerated throughput.
Swarm Multi-Agent Topology
Best Paired With: Gemini 3.8 Flash or Claude 3.7 Sonnet
High code synthesis fidelity for performance-critical systems code and numerical computing.
Executes isolated feature work inside dedicated Git worktrees.
What Developers & Practitioners Are Saying
Real-world field observations, production strengths, and developer consensus
"NVIDIA's ultra-optimized inference model leveraging NVFP4 quantization on Blackwell architectures for extreme throughput."
Verified Real-World Strengths
- Superb inference efficiency and throughput on NVIDIA hardware
- Strong numerical computing, CUDA, and systems programming comprehension
- High mathematical reasoning scores (77.0% AIME)
Production Caveats & Gaps
- Requires modern NVIDIA GPU infrastructure for peak NVFP4 benefits
Primary Sources & Methodology Citations
- NVIDIA Nemotron 3 Ultra Architecture — NVIDIA (January 2026)
Head-to-Head Frontier Comparisons: NVIDIA Nemotron 3 Ultra NVFP4 vs Rivals
Standardized, side-by-side benchmark & economics comparison against peer models
| Model | Role / Tier | SWE-bench | LiveCodeBench | LMSYS Elo | AIME | Input / Output |
|---|---|---|---|---|---|---|
NVIDIA Nemotron 3 Ultra NVFP4Fireworks AICurrent | Tier 2: Accelerated Systems Specialist | 58.2% | 53% | 1365 | 77% | $0.5 / $1.5 |
InklingIndependent Compare NVIDIA Nemotron 3 Ultra NVFP4 vs Inkling → | Tier 2: Code Search & Discovery Specialist | 54.5% | 47% | 1305 | — | $0.15 / $0.6 |
MiniMax-M3MiniMax Compare NVIDIA Nemotron 3 Ultra NVFP4 vs MiniMax-M3 → | Tier 2: Documentation & Test Specialist | 58% | 50.5% | 1320 | — | $0.25 / $1 |
Qwen 3.8 27BQwen Compare NVIDIA Nemotron 3 Ultra NVFP4 vs Qwen 3.8 27B → | Tier 2: Open-Weights Specialist | 62% | 54.5% | 1342 | — | $0.2 / $0.8 |
K2 Horizon 375B A23BIndependent Compare NVIDIA Nemotron 3 Ultra NVFP4 vs K2 Horizon 375B A23B → | Tier 2: MoE Systems Specialist | 59.8% | 52% | 1330 | — | $0.4 / $1.6 |
NVIDIA Nemotron 3 Ultra NVFP4 model overview
NVIDIA Nemotron 3 Ultra NVFP4 is listed in the public Swarm catalog through Fireworks AI. This page summarizes the verified catalog fields used by Swarm without claiming real-time access for every provider account.
- Provider
- Fireworks AI
- Family
- Nemotron
- Context window
- 262.1K tokens
- Thinking support
- Not supported
- Availability status
- Available
- Catalog ID
- fireworks/nemotron-3-ultra-nvfp4
Run NVIDIA Nemotron 3 Ultra NVFP4 in Swarm
Verified CLI workflow & multi-agent architecture
swarm ctl model set --provider fireworks --model fireworks/nemotron-3-ultra-nvfp4# Or switch interactively inside the terminal TUI:
/modelsSubagent Role Matrix
Assign NVIDIA Nemotron 3 Ultra NVFP4 to specialized system roles in ~/.swarm/config.json:
- Primary / Planner: Orchestrates full task DAGs and multi-stage pipelines.
- Coder Subagent: Generates code and unit tests inside private Git worktrees.
- Finder Subagent: Performs read-only code search, symbol discovery, and mapping.
Public catalog snapshot. Catalog availability is not a real-time provider status or an access guarantee.