---
id: "nvidia/nemotron-3-ultra-550b-a55b"
also: ["nvidia/nemotron-3-ultra-550b-a55b-20260604", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "hf/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "requesty/nvidia/nemotron-3-ultra-550b-a55b", "vercel/nvidia/nemotron-3-ultra-550b-a55b"]
cheapest_list_usd_per_m: "0.4 in, 1.76 out, at deepinfra/fp4/flex on openrouter"
cheapest_effective_usd_per_m: "0.1239 in, 1.7516 out, at deepinfra/fp4/flex on openrouter"
context: 262144
max_output: 182520
input: ["text"]
output: ["text"]
released: "2026-06-04"
reasoning: ["high", "medium", "none"]
api: ["/api/v1/chat/completions", "/api/v1/completions", "/api/v1/messages", "/api/v1/responses"]
params: ["frequency_penalty", "include_reasoning", "logit_bias", "logprobs", "max_tokens", "min_p", "presence_penalty", "reasoning", "reasoning_effort", "repetition_penalty", "response_format", "seed", "stop", "structured_outputs", "temperature", "tool_choice", "tools", "top_k", "top_logprobs", "top_p"]
endpoints: 15
gateways: "openrouter (7), vercel (5), requesty (2), hf (1)"
no_training: 13
trains: 1
retains_prompts: 3
intelligence: 11.8000
coding: 49.3000
agentic: 20.1000
---
# NVIDIA: Nemotron 3 Ultra

NVIDIA Nemotron 3 Ultra is an open frontier-reasoning and orchestration model from NVIDIA, with 55B active parameters out of 550B total (MoE). Built on a hybrid Transformer-Mamba mixture-of-experts architecture, it...