Hyperbolic

What it is for

Hyperbolic is an on-demand GPU + inference cloud: you deposit credits, launch an H100 / H200 / B200 instance, or call a serverless model. Use it when the job is rent a listed GPU-hour, not when you need a reserved 1-Click Cluster from one US operator.

It is not open source as a product. On-demand H100 SXM starts at $3.19 / GPU-hour. GPU launches need a $5 deposit (Pro).

What makes it unique: the cover is Hyperbolic’s official on-demand vs reserved vs private-cloud billing table from the docs — hourly vs prepaid vs contract on the same page.

Key features

• On-demand GPU instances — launch H100 / H200 / B200 / B300 boxes from the app for a job you tear down when done (Marketplace; On-demand).
• Serverless inference — call 25+ open models through an OpenAI-shaped chat API without renting a box (Inference).
• Private Cloud clusters — custom topology, networking, and a handover document for production-scale GPU islands (Private Cloud).
• Persistent storage volumes — attach 100 GB–10 TB network disks that survive terminate and can move between instances (Storage).
• Managed Kubernetes — Hyperbolic runs the control plane and NVIDIA operators; you pick GPU workers and get SSO kubectl (Managed Kubernetes).
• Console GPU monitoring — utilization, memory, and temperature per GPU on the instance page, no agent to install (GPU Monitoring).

Who it's for

Best fit: engineers who will deposit $5 and launch a listed GPU-hour.

Poor fit: teams that want a single-operator 8× H100 SKU with a published 1-Click Cluster (Lambda).

Pricing

Verified 2026-10-22 (Asia/Shanghai) from hyperbolic.ai/marketplace, On-demand pricing, and Billing. Marketplace copy: rates refresh weekly. Re-check the row.

On-demand GPU (marketplace “starting from”)

• H100 SXM — $3.19 / GPU / hr
• H200 — $3.99 / GPU / hr
• B200 — $5.99 / GPU / hr

Credits are 1:1 dollars, never expire, $5 minimum purchase. On-demand needs a balance that covers at least one hour of every pending + running instance. Inference: Basic 60 req/min on a free account; Pro 600/min after the $5 deposit. Text tokens are documented from $0.10 / 1M for small models — the live card may list only a few chat models (Llama 3.3 70B / Qwen3-Coder 480B at $0.40 / 1M on a late-2026 shrink of that page).

Limitations / costs

• $3.19 is “starting from” and weekly. Copy the live offer.
• No GPU launch on a $0 balance. Inference may run on Basic; instances need Pro ($5).
• Serverless model list has shrunk. Do not assume every 2025 row still exists.
• Not a Lambda 1-Click Cluster. Networking and support are marketplace-grade.

Alternatives

• Runpod — published Secure Cloud rows; H100 PCIe $2.89/hr.
• Nebius — H100 NVLink $4.50/hr if you want a listed cloud SKU.
• Together AI — open-weight APIs if you will not rent the GPU.

Official tutorial

Official Hyperbolic channel exists. We did not oEmbed a current first-party “open the console / launch an instance” tutorial, so nothing is embedded. Use the written Getting started plus On-demand quick start.

Website

https://www.hyperbolic.ai

Source

Catalog row curated for AI Tools Nav. Cover is Hyperbolic’s official on-demand / reserved / private-cloud billing table. H100 SXM from $3.19/hr. Official channel: @hyperbolic-labs (no getting-started embed).

Last verified

2026-10-22 (Asia/Shanghai) editorial check.