Nemotron 3 Ultra
Open in the interactive archive →
Overview
550B-parameter sparse MoE with 55B active, built for long-running agentic harnesses (OpenCode, Hermes, OpenClaw): hybrid Mamba/Transformer...
Details
- Vendor
- NVIDIA
- Type
- LLM
- Release year
- 2026
- Jurisdiction
- USA
- CEO
- Jensen Huang
- Key staffers
- Jensen Huang (CEO & Founder), Bryan Catanzaro (VP Applied Deep Learning)
- Deployment
- Open weights (Hugging Face, BF16 + NVFP4) with training data, recipes and GenRM reward model
- Open source
- Yes
- Free tier
- Yes
- Stock status
- $131
- X
- @nvidia
Known safety issues
None publicly documented
AIDb Rating
AIDb v2.0 (curated): Capability 8/10 · Currency 10/10 · Access 8/10 · Significance 7/10 → 8.25
Last verified 2026-08-09 · Primary source ↗
Cite this
“Nemotron 3 Ultra.” AIDb — The AI Database. Verified 2026-08-09. https://theaidatabase.com/m/nvidia-nemotron-3-ultra.html
More from NVIDIA
Instant-NGP · Nemotron-4 · Nemotron Nano 8B · Cosmos 1.0 · Cosmos Nemotron 34B · Llama-3.1-Nemotron-Ultra-253B · Llama-3.3-Nemotron-Super-49B-v1 · Nemotron 3.5 ASR