Nvidia

NVIDIA: Nemotron 3 Ultra

nvidia/nemotron-3-ultra-550b-a55b

NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE).

  • Context window: 262,144 tokens
  • Input: text
  • Output: text
  • Pricing: $0.6/M input tokens, $2.4/M output tokens

Community sentiment

Mixed (+12) — 8 positive, 6 negative, 3 neutral across 17 community samples from X and the web, as of 2026-10-30.

Praise centers on the model’s direct, non-sycophantic personality, reasoning, and ability to handle a roughly one-million-token prompt. Criticism points to weaker benchmark results than newer or smaller models and a reported 96-second response time in one comparison.

  • “Interesting, Nemotron 3 Ultra was one of my fav personalities in a model. It wouldnt beat around the bush or be a sycophant. https://t.co/2qJBtXsOC7 https://t.co/Gozl1lkY0I” — X
  • “Smaller beat bigger: Nvidia's distilled Nemotron 3.5 SuperVL reportedly scored 63.44% on HermesBench vs. 56.33% for Nemotron 3 Ultra, with faster generation and multimodal input. See why:… #AIagents https://t.co/PN0Q8sqvL3” — X
  • “OpenCode lists 10 free AI coding models, no sign-in. We asked all 10 the same question. Beam Free answered in 0.4 seconds. Nemotron 3 Ultra took 1 minute 36. Our Oct 18 times, and free models are limited-time. Which one would you bet your project on? https://t.co/OifOokBjhU” — X
  • “I sent a ~1M-token prompt to 8 free models on OpenCode Zen. 2 of 8 took it: Nemotron 3 Ultra and Beam. Flint said it out loud: 764k is longer than its 262k window. The rest just rejected it.” — X

RuntimeWire coverage

View on OpenRouter. Model data sourced from OpenRouter.