IonRouter

IonRouter is a high-performance AI inference platform built by Cumulus Labs, powered by their custom IonAttention engine that multiplexes models on a single GPU with millisecond swap times. It delivers significantly higher throughput than standard inference providers — over 7,000 tokens/second on a single GH200 for Qwen2.5-7B — while keeping costs low with per-token, usage-based pricing. The platform is OpenAI API-compatible, meaning teams can switch with a single line of code change, and supports a wide range of model types including language, vision, image, video, and audio.

• IonAttention Engine: A custom inference stack built from the ground up for NVIDIA Grace Hopper Superchips, enabling real-time model multiplexing and adaptive traffic handling for maximum throughput.
• OpenAI-Compatible API: Point any existing OpenAI client (Python, TypeScript, Go, etc.) at api.ionrouter.io/v1 with no other code changes required.
• Broad Model Support: Access frontier and open-source models across language (Qwen3.5, GPT-OSS-120B, Kimi-K2.5), vision (VLMs), image (Flux Schnell), video (Wan2.2), and audio categories.
• Custom Model Deployment: Deploy your own finetunes, custom LoRAs, or any open-source model on dedicated GPU streams with no cold starts and per-second billing.
• Usage-Based Pricing: Pay per million tokens (input and output priced separately) or per GPU-second for video/image generation — no idle costs or seat fees.
• Iondex Leaderboard: Browse and compare available models via the Iondex leaderboard and category explorer to find the best model for your use case.
• Playground: Test any supported model directly in the browser playground before integrating into your application.
• Enterprise Support: Book a call for dedicated GPU streams, custom deployments, and enterprise-grade SLAs tailored to high-volume workloads like robotics, surveillance, and AI video pipelines.