KymaKyma
NVIDIA

NVIDIA

Nemotron 3 Ultra 550B

NVIDIA's open-weight model optimized for reasoning, coding, and long-context text generation. Engineers should use it when they need a balance of throughput and frontier-level performance without proprietary API constraints.

Modalities

Text → Text

Below release

fp4 · fp8

Weights

Published

nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16

Where it sits in the catalogue Kyma measures

Every number against every text model Kyma prices per token — a stated rule, not a chosen line-up.

Price · input#30/70

$0.675

output $3.375 · cached $0.135

median $0.977 · best $0.0494

Price · output#33/70

$3.375

per 1M tokens generated

median $3.4515 · best $0.2164

Availability · 30d#10/70

98.78%

411 observations since 2026-06-10

median 93.7% · best 100.0%

Throughput#29/70

46.1 tok/s

probe median, one fixed prompt

median 39.4 tok/s · best 314.6 tok/s

Response time#23/70

1.71 s

probe median, to a complete answer

median 2.53 s · best 0.84 s

Context

1M

max output 32,768

published by NVIDIA, not measured here

Tick above each rail is this model, below it the other 69. Dashed rule is the field median, solid is its best. Better is left; the axis stops at the 90th percentile, so a few models sit past its right edge.

Usage

How much this model is actually called here.

Rank

#52

of 117 active models

Tokens served

452.7K

all-time

Platform share

0.0%

of all tokens

Pricing

Pay per token. Cached input bills at this model’s own cached rate, listed below.

$0.675 /1M input$3.375 /1M output
Hobby10 req/day · 2K in / 500 out
~$0.91/mo
Production1,000 req/day · 2K in / 500 out
~$91.13/mo
Scale20,000 req/day · 2K in / 500 out
~$1,823/mo
+ Estimate your workload
1,000
2,000
500
30%

Estimated monthly cost

$81.41

$2.71 / day on Nemotron 3 Ultra 550B

Same workload on:

Gemini 3.8 Flash$273+236%
Claude Fable 5.1$1,823+2139%

Estimates use list pricing with cached input at this model's own cached rate. Actual bills depend on real token counts, and every response includes its exact cost.

When to use Nemotron 3 Ultra 550B

Updated 2026-07-31

Where this model earns its cost — and where it doesn't.

Nemotron 3 Ultra 550B is a 550B parameter mixture-of-experts architecture with 55B active parameters per forward pass. It combines a hybrid Mamba-Transformer design to handle reasoning, coding, and general-purpose text tasks efficiently.

On Kyma, the model runs through an OpenAI-compatible endpoint with automatic request failover and exact cost reporting in the response payload. It supports tool use, structured outputs, and prompt caching, which bills repeated prefixes at this model's cached input rate.

The model accepts a 1,000,000 token context window but caps generation at 32,768 output tokens. It is strictly text-only and does not support vision or multimodal inputs.

Long Context Document Analysis

Process large technical manuals or codebases within the one million token window.

Automated Code Generation

Generate and refactor code using built-in reasoning and tool-use capabilities.

High Throughput Chat

Handle conversational workloads that require fast text generation and structured JSON outputs.

Reasoning Heavy Workflows

Execute multi-step logical tasks where active parameter routing improves accuracy.

Not ideal for: Tasks requiring image or video understanding, or workloads that need output sequences longer than 32,768 tokens.

How it compares

Against the peers people actually weigh it against.

SpecNemotron 3 Ultra 550BGemini 3.8 FlashClaude Fable 5.1
Input /1M$0.675$2.026$13.50
Output /1M$3.375$10.126$67.50
Context1M1M1M
ToolsYesYesYes
ReasoningYesYesYes
Throughput46.1 tok/s34.7 tok/s12.7 tok/s

Quick start

Up and running in under two minutes.

  1. 1

    Create an API key

    Sign up and grab a key from the dashboard. Nemotron 3 Ultra 550B needs a top-up — the signup credit covers the free tier.

    Get API key →
  2. 2

    Make your first request

    Drop in your key and send a chat completion — fully OpenAI-compatible.

    curl https://kymaapi.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nemotron-3-ultra-550b",
    "messages": [
      {"role": "user", "content": "Explain prompt caching in one paragraph."}
    ]
  }'
  3. 3

    Stream responses

    Add "stream": true to receive tokens as they arrive.

    curl https://kymaapi.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nemotron-3-ultra-550b",
    "stream": true,
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

FAQ

Common questions about this model.

What is the context window of Nemotron 3 Ultra 550B?

Nemotron 3 Ultra 550B has a 1M-token context window — roughly 1471 pages of text in a single request.

How much does the Nemotron 3 Ultra 550B API cost?

$0.675 per 1M input tokens and $3.375 per 1M output tokens, with cached input at $0.135/1M — 80% off the input rate on repeated prompt prefixes. No subscription; you pay only for what you use.

Does Nemotron 3 Ultra 550B support function calling?

Yes — Nemotron 3 Ultra 550B supports tool/function calling and structured outputs (JSON mode), so it works with agent frameworks out of the box.

Are the weights for Nemotron 3 Ultra 550B publicly available?

Yes. NVIDIA publishes Nemotron 3 Ultra 550B's weights as nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16, so you can download and run the model yourself (https://research.nvidia.com/labs/nemotron/Nemotron-3-Ultra/, read 2026-08-14). Kyma serves it because it is convenient and has failover behind it, not because it is the only way to reach it.

Is Nemotron 3 Ultra 550B ever served below the precision its creator released it at?

Sometimes. Nemotron 3 Ultra 550B was released by NVIDIA at bf16 (https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16, read 2026-08-14), and at least one route serving it here reports fp4 or fp8 — narrower than that. Weights compressed below the release usually answer close to it, but it is not the identical artefact, and nothing in the response tells you which one answered. Kyma checked the routes on 2026-09-03, so it is your call rather than a silent one.

How do I use Nemotron 3 Ultra 550B?

Kyma is OpenAI-compatible: point your SDK's base URL at https://kymaapi.com/v1, use your Kyma API key, and set the model to nemotron-3-ultra-550b. Signing up is free, but this model needs a top-up: the $0.50 signup credit covers the free tier only.

Does this model support image inputs?

No, it is strictly text-only for both input and output modalities.

How does prompt caching work with this model?

Repeated prompt prefixes are billed at this model's cached input rate, reducing costs for long-context or system-prompt-heavy workflows.

Can I use function calling and structured outputs?

Yes, the model natively supports tool use and structured JSON responses, which are routed through Kyma's OpenAI-compatible API.

Needs a top-up — the signup credit covers the free tier.Create account →