NVIDIA Nemotron 3 Ultra 550B-A55B
Nemotron 3 Ultra is a reasoning model from NVIDIA, released on 2026-06-04. It accepts text input and returns text output. The cataloged parameter count is 550B, with 55B active parameters per inference. The recorded context window is 1M. Cataloged capabilities include Reasoning model and Multilingual. The checkpoint is listed under the Free Commercial license. Use the linked references to confirm current access, licensing, and provider-specific limits.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
Nemotron 3 Ultra currently shows benchmark results led by IF Bench (4 / 282, score 81.70), IMO-AnswerBench (1 / 24, score 92.30), Pinch Bench (2 / 38, score 90). This page also consolidates core specs, context limits, and API pricing so you can evaluate the model from benchmark results and deployment constraints together.
Nemotron 3 Ultra is a reasoning model from NVIDIA, released on 2026-06-04.
It accepts text input and produces text output. Its cataloged capabilities include Reasoning model and Multilingual. The cataloged parameter count is 550B, with 55B active parameters per inference. The recorded context window is 1M.
The checkpoint is listed under the Free Commercial license. The evaluation section contains 2 cataloged benchmark results with their recorded modes and scores. The page links 4 release, model-card, repository, or provider references for checking the underlying claims. Specifications, availability, and prices can change; undisclosed values are intentionally left unstated.
Nemotron 3 Ultra is a reasoning model from NVIDIA, released on 2026-06-04. It accepts text input and returns text output. The cataloged parameter count is 550B, with 55B active parameters per inference. The recorded context window is 1M. Cataloged capabilities include Reasoning model and Multilingual. The checkpoint is listed under the Free Commercial license. Use the linked references to confirm current access, licensing, and provider-specific limits.
The current model record lists text as input and text as output.
The cataloged parameter count is 550B, with 55B active parameters per inference. The recorded context window is 1M. Fields without a source-backed value remain undisclosed.
Open-weight checkpoint; NVIDIA does not publish a model-specific first-party token price for self-hosting this checkpoint.
The evaluation section contains 2 cataloged benchmark results with their recorded modes and scores. Compare only results that use the same benchmark version and evaluation mode.
The checkpoint is listed under the Free Commercial license. Review the linked license text before commercial or derivative use.
Follow DataLearner on WeChat for AI model updates and research notes.
