Expired50% confidenceBenchmarkExact time
ExoLabs can already run the 550-billion-parameter Nemotron 3.0 on 4 Spark units at 30 tokens per second.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/19/2026
First Seen
Valid until: 10/17/2026(expired)
Sources
本地AI爆发拐点：端侧部署、多模型协作与开源生态全解析
Related Claims
UnverifiedNVIDIA released Nemotron 3 Ultra, featuring 550 billion parameters and a MoE architecture, designed specifically for long-running AI agent tasks.68% similarUnverifiedIn the Nemotron model series, "Nano" in the name indicates a parameter count in the billions range (typically 4B–8B parameters), suitable for running on a single consumer-grade GPU.63% similarVerifiedThe M4 chip's Neural Engine can perform 38 trillion operations per second (38 TOPS)61% similarUnverifiedNVIDIA has developed a model called Nemotron Ultra with 550 billion parameters61% similarUnverified去噪塔仅用2.1万亿token训练，而原始Nemotron Nano使用了25万亿token，因此该模型是解码方式的转换而非更聪明的模型61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/561418API
curl https://kongchang.com/api/v1/knowledge/claims/561418MCP
get_claim(id=561418)