NVIDIA Nemotron 3 Ultra

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

Run NVIDIA Nemotron 3 Ultra in a hosted Python Task with extended thinking and file inputs.

NVIDIA's Nemotron 3 Ultra via OpenRouter. An open reasoning model on a hybrid Transformer-Mamba architecture (550B total, 55B active parameters) with a 262K token context window.

What NVIDIA Nemotron 3 Ultra supports in qbash

Extended thinkingReasons through a problem before it replies.
Reasoning effort2 levels, from medium to high.
Document inputReads PDF, TEXT, DOC, XLS, RTF and 7 more files.
Prompt cachingRepeated input is billed at the cached rate.

What is qbash?

qbash is where you build Tasks: Python programs that run a business process on managed infrastructure. Your code runs the steps and calls a model like NVIDIA Nemotron 3 Ultra only at the ones that need judgment, with integrations, memory and files available in the same script.

A developer writes the Task once, and anyone on the team runs it from a form, on a schedule, from a webhook or from chat. Each Task is versioned and every run is logged.

How Tasks work

Specifications

Context window: 262Ktokens
Max output: 131,072tokens
Temperature: 0 to 1default 1

Use NVIDIA Nemotron 3 Ultra in a Python Task

qbash.ai.raw calls NVIDIA Nemotron 3 Ultra from a Task and returns its reply as a string.

main.py
notes = qbash.inputs.get("notes")

summary = qbash.ai.raw(
    f"Summarize these meeting notes in five bullets:\n\n{notes}",
    model="nvidia/nemotron-3-ultra-550b-a55b",
    system="You are a concise analyst.",
)

qbash.output({"summary": summary})

Model calls in Tasks

Reasoning effort with NVIDIA Nemotron 3 Ultra

Set Effort to medium and high in the prompt's Settings. Higher effort spends more tokens on reasoning before the reply.

Turn on Thinking in the same Settings popover to have NVIDIA Nemotron 3 Ultra reason through a problem before it replies.