PRICING

Dedicated and
serverless endpoints

DEDICATED ENDPOINTS

Reserve dedicated GPUs

Your model on GPUs reserved for you.

  • Throughput and latency SLA
  • Tuned for your traffic
  • Catalog models, your own 
    models and fine-tunes
Talk to an engineer
SERVERLESS ENDPOINTS

Start on shared GPUs

No capacity to plan or reserve.

  • Pay for the tokens you use
  • 11 popular open models
  • Other models priced by size
View token prices

Pricing

DEDICATED ENDPOINTS

GPU pricing

GPUs reserved for your endpoint.

From$1.10per GPU-hour

Per month assumes 730 hours, one GPU running around the clock.

GPUMEMORYGOOD FORLIST / HOUR
L40S48 GBSmall models, embeddings and vision$1.10
RTX PRO 600096 GBMid-size models on one GPU$1.25
H10080 GBGeneral production LLM serving$3.15
H200141 GBLarger models and longer context$3.80
B200180 GBHigh-throughput serving$5.95
B300288 GBThe largest open models$6.65
GB300 NVL72288 GB per GPUVery large models across 72 GPUsPriced per deal
SERVERLESS ENDPOINTS

Token pricing

Price per 1M tokens on shared GPUs.

From$0.0125per 1M input tokens
MODELINPUTOUTPUT
Kimi K3moonshotai/Kimi-K3$0.75$3.75
GLM 5.3zai-org/GLM-5.3$0.35$1.10
NVIDIA Nemotron 3 Ultranvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16$0.15$0.60
DeepSeek R1deepseek-ai/DeepSeek-R1$0.15$0.60
MiniMax M3MiniMaxAI/MiniMax-M3$0.075$0.30
DeepSeek V4 Flashdeepseek-ai/DeepSeek-V4-Flash$0.075$0.30
GLM 4.6zai-org/GLM-4.6$0.075$0.30
Qwen3 235B InstructQwen/Qwen3-235B-A22B-Instruct-2507$0.075$0.30
gpt-oss-120bopenai/gpt-oss-120b$0.04$0.15
NVIDIA Nemotron 3.5 Lightning 30Bnvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16$0.0125$0.05
Qwen3 Coder 30BQwen/Qwen3-Coder-30B-A3B-Instruct$0.0125$0.05
SERVERLESS ENDPOINTS

Pricing by model size

For open models not listed above.

From$0.025per 1M tokens, input and output
MODEL SIZEINPUT AND OUTPUT, PER 1M TOKENS
Dense, under 4B$0.025
Dense, 4B to 16B$0.05
Dense, over 16B$0.225
Mixture of experts, up to 56B$0.125
Mixture of experts, 56B to 176B$0.30
Mixture of experts, over 176BOn request

Estimated workload costs

Most dedicated quotes come in below list price.

GPU

ESTIMATED MONTHLY COST AT LIST

$18,396

$25.20 per hour for the whole endpoint

8 × H100 × $3.15 per GPU-hour × 730 hours

We size the GPUs to your traffic.

Talk to an engineer

How Saturn Cloud compares

Lower price per GPU-hour

$5.95Saturn Cloud B200 list price

Dedicated inference

List price, $ per GPU-hour

More tokens per GPU

2.4xGLM-5.3 throughput after tuning

GLM-5.3, 16 live sessions

Tokens per second per GPU, same four GB300 GPUs

List prices from each provider’s public pricing page.

What comes with every dedicated endpoint

Performance

  • Workload profiling and tuning
  • Throughput and latency SLA
  • Built on NVIDIA Dynamo

Models

  • Open models from our catalog
  • Your own models
  • Fine-tuning on your data, on models that support it

Deployment

  • GPUs from our neocloud partners
  • Your own GPUs or cloud account
  • Air-gapped networks

Security

  • SOC 2
  • GPUs dedicated to your endpoint
  • A specific region for data sovereignty

How it works

  1. 01

    Tell us the workload

    Your model, traffic pattern, latency target and region.

  2. 02

    Get a price

    We match GPUs and a region to the workload and send a quote.

  3. 03

    Go live with an SLA

    We deploy, tune and benchmark the endpoint, 
    then commit to the measured performance.

FAQ

Pricing questions

Why is GPU capacity quoted instead of self-serve?

The capacity comes from neocloud partners and isn’t in a public catalog, so we quote each workload.

Can I bring my own model?

Yes. Serve a model from our catalog, bring your own, or fine-tune one on your data and serve the result from your dedicated endpoint.

Can I try it before I buy?

Yes. Sign up for a free account to try the platform. When you’re ready for dedicated GPU capacity, talk to us.

Why are your prices lower than other inference providers?

We partner with neoclouds bringing new GPU capacity online, before it reaches the open market. We also tune each deployment, so you get more tokens from every GPU-hour.

Can I run on my own hardware or in my own cloud account?

Yes. The platform only needs a Kubernetes cluster, so it can run on-premises, in your cloud account, in air-gapped networks or in a specific region for data sovereignty.

I already have a self-service account. 
Where are the hourly rates?

Hourly rates for self-service accounts are on the Saturn Cloud Hosted page.

Get a price for your workload

Send us your model and traffic. We’ll come back with a quote and the performance we’ll commit to.