Singapore, September 10, 2026 - Ocean Network, the peer-to-peer GPU cloud for AI where inference and compute jobs are booked by the hour and yours alone while they run, today announced the launch of Inference, enabling users to deploy AI models and applications as persistent, HTTP-accessible services on dedicated GPU hardware. The launch represents the platform’s largest capability expansion to date.
Ocean Network Inference is designed to resolve three recurring friction points in the AI infrastructure market: unpredictable billing, lack of model transparency, and hardware uncertainty. Independent community research has documented widespread concern about inference providers substituting or routing models without user visibility, alongside billing surprises on long-running workloads.
Ocean Network addresses these through its architecture. The platform displays the full session cost before any funds are committed, bills by the hour rather than by the token, and routes each request to a dedicated node rather than shared infrastructure. The model selected by the user is the model that runs.
“We did the research, read the threads, and built what the community kept asking for: a normal endpoint, a normal card, and a price you know before you start. Compute jobs gave users the tools to build and test. Inference lets them run. That is the full circle, and we have been building toward it since day one,” says the Ocean Network team.
The pricing for an H200 GPU on Ocean Network starts at $2.16 per hour. In internal testing, a 30 second video generated using the platform’s Templates feature was produced in approximately 20 minutes at a cost of $0.72. Payment is handled on-chain through an escrow contract on Base, with the node operator claiming the full booked session window before the container starts.
Ocean Network Inference offers four entry points, differing only in how much configuration is handled by the user versus the platform:
- Curated Models: Pre-configured model and hardware packages for immediate deployment, including Qwen3-8B and large reasoning models across GPU configurations from 16GB single-card to multi-card setups.
- Custom Models: Full configuration control for any model on the Hugging Face Hub, running on vLLM or llama.cpp, with an OpenAI-compatible API output.
- Services: Containerised applications published by node operators, including ComfyUI, Open WebUI, JupyterLab, and others.
- Templates: Ready-to-run application and model bundles for specific outcomes, including text-to-image, image-to-video, and reference-to-video generation.
To mark the launch, Ocean Network is making 6 hours of H200 inference available in complimentary tokens. Inference is available now at dashboard.oncompute.ai. Users require a wallet with funds on Base, available through an external wallet or the platform’s built-in Smart Wallet with email or social login.
About Ocean Network
Ocean Network is a decentralized, peer-to-peer GPU compute network that turns idle and underutilized hardware into accessible, pay-per-use compute resources. It lets users choose a node with the resources they need, submit a containerized compute job, and get results back without managing servers or infrastructure. With the launch of Inference, the network now also supports long-running AI model deployments and applications, reachable over HTTP for as long as the session is booked.
