FrontierJune 4, 2026via MarkTechPost
NVIDIA AI Releases Nemotron 3 Ultra: An Open 550B Mixture-of-Experts Hybrid Mamba-Transformer for Long-Running Agents
Why it matters
NVIDIA's Nemotron 3 Ultra challenges the closed-model dominance in long-context reasoning and agentic workloads by shipping an open alternative with superior throughput and a 1M-token window—forcing teams to recalculate their inference cost models.
Key signals
- 550B total parameters (55B active via MoE)
- 1M-token context window
- ~6x higher inference throughput vs. comparable open LLMs
- Parity accuracy with larger closed models
- Open weights + training data + recipes
- OpenMDW-1.1 license
- Hybrid Mamba-Transformer architecture
- Optimized for long-running agents
The hook
550B parameters. 6x faster inference. NVIDIA just open-sourced a model built for agents that changes what's possible at scale.
NVIDIA has released Nemotron 3 Ultra, a 550B total (55B active) open Mixture-of-Experts hybrid Mamba-Transformer for long-running agents. It pairs a 1M-token context with up to ~6x higher inference throughput than comparable open LLMs at on-par accuracy, and ships with open weights, training data, a…