#20NVIDIAOpen weights

Nemotron 3 Ultra 550B A55B

NVIDIA’s open-weight 550B-parameter mixture-of-experts reasoning model, with 55B active parameters and independent Artificial Analysis results.

Official provider page Compare with another model

Overall: 46.1
Input $/M: $0.60
Output $/M: $2.50
Context: 262K
Released: Jun 4, 2026
Updated: Oct 6, 2026

Category scores

Reasoning & Math
rank 21
39.7
Speed
rank 3
71.6
Open-source
rank 8
39.7

Overall rank history

Weekly snapshots. Lower is better.

Strengths

Fast generation at 179.1 output tokens per second
Open weights with 55B active parameters
262,144-token context window

Weaknesses

Below the current frontier leaders on the Intelligence Index
Large 550B total-parameter model requires substantial hardware to self-host