Inference for text, vision, and audio.

Generate text, understand images, and produce speech. Ion runs the models behind your application.

Get model API accessAPI documentation

Hosted on IonRouter. Models & pricing

Loading product visual

Tune model execution

Model-specific kernels and memory management.

Swap models

Load supported models as requests change.

Use your existing SDK

Call Ion through an OpenAI-compatible API.

Hardware, models & resultsInside the Ion engine

Swap models on one GPU.

Finish a request, load the next model, and reuse the GPU.

Run a model.

Ion inference
Cluster 05 · Grace HopperNVIDIA
Language modelLlama
Ready
Switch policy: Drain
  1. Finish request
  2. Load weights
  3. Next request
Text inference

RequestSummarize this report.

ResultCapacity is ready for the next deployment.

Load another model on this GPU.
System RAM
  • Llama
  • Qwen
Model weights
NVLink-C2CWeight transfer
GPU memory
Llama
Loaded weights
1 of 3
Use ← → to explore

Explore models and clusters
Select a cluster9 clusters
Select a model family

From account to first request.

Create an IonRouter account, add credits, and send a request.

Create your account

Send your first request

curl https://api.ionrouter.io/v1/chat/completions \
  -H "Authorization: Bearer $ION_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3-30b-a3b",
    "messages": [{
      "role": "user",
      "content": "Summarize why GPU sharing matters."
    }],
    "max_tokens": 128
  }'
Models & pricingAPI setup guide

Hosted or dedicated inference.

Dedicated GPUs

Cumulus GPUs configured for your models and traffic.

Book a demo

Your infrastructure

Deploy Ion on your hardware. Grace Hopper optimizations need a coherent CPU–GPU link.

Book a demo

Related products

PaladinManage GPU capacityTalosBuild and run agents

Questions

Run your models on Ion.

Get model API accessBook a demo