Perplexity adds NVIDIA Nemotron 3 Ultra
Perplexity has integrated NVIDIA's newly released flagship model, Nemotron 3 Ultra, into its Pro and Max subscription tiers. Nemotron 3 Ultra is a 550-billion parameter hybrid LatentMoE model utilizing 55 billion active parameters, Mamba-2, and Attention layers with a 1-million token context window, optimized specifically for complex planning, deep reasoning, and agentic workflows.
Perplexity's rapid adoption of Nemotron 3 Ultra shows that search platforms must constantly cycle in the newest frontier open models to stay competitive against closed ecosystems.
- –Validates the viability of hybrid Mamba-2 and LatentMoE architectures for massive-scale consumer deployment.
- –Enhances Perplexity Pro/Max subscription value by adding high-throughput, long-context reasoning engines.
- –Illustrates NVIDIA's capability to challenge proprietary models in deep reasoning and multi-agent coordination.
DISCOVERED
2h ago
2026-06-05
PUBLISHED
2h ago
2026-06-05
RELEVANCE
AUTHOR
AravSrinivas