IonRouter

High throughput, low cost inference. Powered by IonAttention.

Not just fast hardware. A faster engine: IonAttention.

Our custom inference stack multiplexes models on a single GPU, swaps in ms, and adapts to traffic in real time. Built from the ground up for Grace Hopper.

Throughput (tok/s)
Single GH200, Qwen2.5-7B
IonAttention
7,167
Top inference provider
~3,000

Bring any model. Get dedicated streams.

Deploy your finetunes, custom LoRAs, or any open-source model on our fleet. Dedicated GPU streams with no cold starts and per-second billing.

Custom Models

GPU Stream

GH200
0ms cold start
api.ionrouter.io/v1

• ⚙Robotics: Real-time VLM perception
• ◉Surveillance: Multi-stream video analysis
• ▦Game Gen: On-demand asset generation
• ▶AI Video Gen: Text & image to video

From robots to real-time video.

Teams use Ion for highest performance robotics perception, multi-camera surveillance, game asset generation, and AI video pipelines.

Case Study

5 VLMs, 1 GPU.

Five vision-language models on a single GPU — 2,700 video clips, concurrent users, <1s cold starts.

API · Zero Code Changes

Point your existing OpenAI client at Ion. Any language, any framework. One line change.

Example:

from openai import OpenAI

client = OpenAI(
    api_key="sk-your-key-here",
    base_url="https://api.ionrouter.io/v1"
)

response = client.chat.completions.create(
    model="qwen3.5-122b-a10b",
    messages=[{
        "role": "user",
        "content": [
            {"type": "image_url", "image_url": {"url": "..."}},
            {"type": "text", "text": "What's in this image?"}
        ]
    }]
)

Models & Pricing

Pay per million tokens. No idle costs.

• GLM-5
  ZhiPu AI's flagship 600B+ MoE model with state-of-the-art reasoning, coding, and multilingual capabilities, powered by EAGLE speculative decoding on 8x B200 GPUs.
  ~220 tok/s
  $1.20 in · $3.50 out
  Try in Playground glm-5

• Kimi-K2.5
  MoonShot AI's frontier reasoning model designed for long document understanding, multi-step reasoning chains, and complex problem decomposition across technical and scientific domains.
  ~120 tok/s
  $0.20 in · $1.60 out
  Try in Playground kimi-k2.5

• MiniMax-M2.5
  MiniMax's flagship 1M-context language model delivering strong reasoning and instruction following across long documents, multi-turn dialogue, and complex analysis.
  ~120 tok/s
  $0.40 in · $1.50 out
  Try in Playground minimax-m2.5

• Qwen3.5-122B-A10B
  Cumulus's most capable open-source model — a 122B MoE with 10B active parameters rivaling leading proprietary models on coding, reasoning, and multilingual benchmarks.
  ~120 tok/s
  $0.20 in · $1.60 out
  Try in Playground qwen3.5-122b-a10b

• GPT-OSS-120B
  A frontier open-source 120B model delivering cutting-edge reasoning and instruction following comparable to leading closed-source systems, ideal for complex agentic workflows and advanced code generation.
  ~100 tok/s
  $0.020 in · $0.095 out
  Try in Playground gpt-oss-120b

• Wan2.2 Text-to-Video
  A 14B text-to-video model optimized for speed via the FastGen runtime, generating clips in under 10 seconds with strong motion coherence.
  ~8s/clip
  $0.00194 / GPU·sec
  Try in Playground wan2.2-t2v-general

• Flux Schnell
  Black Forest Labs' fastest Flux model, delivering crisp sub-4-second image generation ideal for real-time applications, prototyping, and high-volume pipelines.
  ~3s/image
  ~$0.005 in · per image out
  Try in Playground flux-schnell

Ready to build?

Start in under a minute. No GPU expertise required.