SaaufAI
  • Loading…
  • WispsAssistants that keep working while you're awayPage
  • DocsGuides with screenshotsPage
  • Quick startDocs · first Wisp or first API callPage
  • API referenceDocs · Chat Completions, streaming, toolsPage
  • Create a WispDocs · step by stepPage
  • BlogNews and notesPage
  • All modelsDirectoryPage
  • CompaniesWho makes the modelsPage
  • Compare disclosuresEvery model, every questionPage
  • PoliciesTerms, privacy, acceptable usePage

Nemotron 3 Ultra

NVIDIA's largest open Nemotron 3 model (550B, 55B active) for reasoning, coding and agents.

Availableby NVIDIA
Use this model Compare
Input · 1M
€0.537
Output · 1M
€2.15
Context
202.8K
Max output
182.5K
Tokens / s
90
Disclosed
4/10
01

Disclosures

3 yes1 no6 open
  1. 01
    Are prompts and outputs stored by SaaufAI?

    Whether this gateway keeps the content of requests or responses.

    Not specified
  2. 02
    May the upstream provider use data for model training?

    Based on the upstream provider's API terms for this model.

    Not specified
  3. 03
    Does the upstream provider retain data for abuse monitoring?

    Many providers keep API data for a limited period (often up to 30 days).

    Not specified
  4. 04
    Is data processed outside the EU/EEA?

    Whether requests may be routed to servers outside the EU/EEA.

    Not specified
  5. 05
    Is the model suitable for personal or sensitive data?

    Operator recommendation for sending personal data to this model.

    Not specified
  6. 06
    Does the model support image input?

    Whether you can send image_url content parts. Answered automatically from the model's capabilities.

    No
  7. 07
    Does the model support tool / function calling?

    Whether the tools parameter is supported. Answered automatically from the model's capabilities.

    Yes
  8. 08
    Is the model output content-filtered by the provider?

    Whether the upstream provider applies its own safety filters.

    Not specified
  9. 09
    Does SaaufAI enforce its own safeguards on this model?

    Usage limits or content rules checked by SaaufAI before a request reaches the provider. Answered automatically from the live configuration.

    Active: prompts screened for sexual content involving minors; prompts screened for cybersecurity, fraud & financial crime, illegal drug production, weapons & explosives, chemical & biological threats, violent extremism, self-harm & suicide, harassment & doxxing, adult sexual content, personal & payment data, jailbreaks & prompt injection. Prompts are screened by GPT 6 Luna Decisions, Gemini 3.5 Flash-Lite (via Google), D3; the screening model sees their content.

    Yes
  10. 10
    May requests be answered by a different (fallback) model?

    If the model fails or a safeguard is reached, the request can be served by another model without the response saying so, possibly at a different provider with different data terms. Answered automatically.

    If the model is unavailable: Grok 4.20 Fast, Gemini 3.5 Flash-Lite, Gemini 3.1 Flash-Lite.

    Yes

Based on the upstream provider's terms. Yes or No is not a rating. how to read these.

02

Pricing

TokensPer 1MPer 1K
InputPrompt tokens€0.537€0.000537
Cached inputPrompt-cache hits€0.107€0.000107
OutputGenerated tokens, reasoning included€2.15€0.00215

What requests cost

  • Short message1K in · 500 out€0.00161
  • Long prompt100K in · 2K out€0.058
  • 1,000 short messages1K in · 500 out each€1.61

Prices in EUR, charged per token. Spend counts toward your key's spending limit, if it has one. Cached input applies when the provider reports prompt-cache hits.

03

Capabilities

Speed
90 tokens/s Fast
First token after 2.4 s · median of streamed answers, last 24 hours
Input
  • Text: supported
  • Image: not supported
  • Audio: not supported
  • Video: not supported
  • Files (PDF…): not supported
Output
  • Text: supported
  • Image: not supported
  • Audio: not supported
  • Video: not supported
  • Embeddings: not supported
  • Transcription: not supported
  • Decisions: not supported
Features
  • Streaming: supported
  • Function calling: supported
  • Structured outputs: not supported
  • Reasoning: supported
Endpoints
  • /v1/chat/completions: supported
  • /v1/responses: not supported