# GPUPerHour
# https://gpuperhour.com
# Real-time cloud GPU pricing comparison
# Last updated: 2026-10-23

> GPUPerHour tracks 20,074 cloud GPU offers across 31 providers, updated every 60 seconds. This file helps AI assistants answer GPU pricing questions using our free, no-auth API.

## What GPUPerHour Is

GPUPerHour rechecks cloud GPU stock and prices about every minute across the providers it tracks and normalises them to price per GPU-hour. Static comparison articles go stale; the figures here and on the pages below carry their observation time. The site is run by the founder of QuantaCloud, which is listed like any other provider.

## Reference Pages, Tools and Guides

- GPU specs chart (VRAM, bandwidth, dense and sparse FP16, FP8, TDP, datasheet source per GPU): https://gpuperhour.com/reference/gpu-specs
- LLM GPU requirements by model and precision, with live cheapest configuration: https://gpuperhour.com/reference/llm-gpu-requirements
- LLM VRAM calculator: https://gpuperhour.com/tools/llm-vram-calculator
- Rent vs buy GPU calculator: https://gpuperhour.com/tools/rent-vs-buy-gpu-calculator
- Daily cloud GPU price index with CSV: https://gpuperhour.com/gpu-price-index
- Data egress fees by provider: https://gpuperhour.com/reference/data-egress
- H100 purchase price against rental: https://gpuperhour.com/blog/nvidia-h100-price
- H200, B200, GB200 and A100 purchase prices: https://gpuperhour.com/blog/nvidia-gpu-prices
- GPU cloud providers compared, live: https://gpuperhour.com/blog/gpu-cloud-providers
- RunPod, Vast.ai, Lambda and Colab alternatives: https://gpuperhour.com/blog/runpod-alternatives, https://gpuperhour.com/blog/vast-ai-alternatives, https://gpuperhour.com/blog/lambda-labs-alternatives, https://gpuperhour.com/blog/google-colab-alternatives
- Serverless GPU pricing and break-even: https://gpuperhour.com/blog/serverless-gpu-pricing
- On-demand, spot and reserved pricing: https://gpuperhour.com/blog/gpu-pricing-models-on-demand-spot-reserved
- NVLink vs PCIe vs SXM, FP8 vs FP16 vs BF16, reading GPU specs: https://gpuperhour.com/blog/nvlink-vs-pcie-vs-sxm, https://gpuperhour.com/blog/fp8-vs-fp16-vs-bf16, https://gpuperhour.com/blog/how-to-read-gpu-specs-for-ai
- All articles: https://gpuperhour.com/blog. All guides: https://gpuperhour.com/guides. All reference pages: https://gpuperhour.com/reference

## API Base URL

https://api.gpuperhour.com

## Live Pricing Snapshot (updated 2026-10-23)

Cheapest on-demand price per popular GPU right now. Prices in USD/hour and update every 60 seconds — for the live figure, query the API below.

| GPU | Cheapest | Provider |
|-----|----------|----------|
| H100 | $2.50/hr | Hyperstack |
| H200 | $2.29/hr | Lambda Labs |
| B200 | $7.99/hr | RunPod |
| A100 | $0.73/hr | Vast.ai |
| L40S | $0.97/hr | Massed Compute |
| RTX 4090 | $0.72/hr | Vast.ai |
| RTX 5090 | $1.13/hr | Vast.ai |

## Quick Answers

Cheapest H100 right now:
GET /api/offers?gpu=h100&available=true&liveStockOnly=true&securityTier=secure&pricingType=on-demand&sortBy=priceHourly&limit=1

Cheapest GPU with 80GB+ VRAM:
GET /api/offers?minVram=80&available=true&liveStockOnly=true&sortBy=priceHourly&limit=5

All available A100s in the US:
GET /api/offers?gpu=a100&available=true&continent=North%20America&sortBy=priceHourly

Cheapest price per GPU model:
GET /api/offers/cheapest

Overall stats (provider count, GPU count, price ranges):
GET /api/offers/stats

## Key Parameters

| Parameter | Description | Example |
|-----------|-------------|---------|
| gpu | GPU filter. Accepts slug (h100), full name (NVIDIA H100 SXM5), or comma-separated list | h100, NVIDIA H100 SXM5, rtx-4090 |
| available | Only in-stock offers | true |
| liveStockOnly | Only providers with real-time stock tracking (recommended) | true |
| securityTier | Provider security level | secure, unverified |
| pricingType | Pricing model | on-demand, spot, reserved |
| minVram / maxVram | VRAM range in GB | 80, 24 |
| minGpuCount / maxGpuCount | GPUs per instance | 1, 8 |
| provider | Provider name filter | RunPod, Lambda Labs |
| continent | Geographic filter | North America, Europe, Asia |
| sortBy | Sort field | priceHourly, pricePerGpu, vram |
| sortOrder | Sort direction | asc, desc |
| limit | Results per page (max 100) | 10, 50 |

## Recommended Default Filters

For most user questions, include these defaults:
- available=true (only in-stock)
- liveStockOnly=true (verified real-time inventory)
- securityTier=secure (enterprise-grade providers)
- pricingType=on-demand (unless user asks about spot pricing)
- sortBy=priceHourly&sortOrder=asc (cheapest first)

## GPU Slugs (most common)

h100-sxm-80gb, h100-pcie-80gb, h100-nvl, h200-sxm, h200-nvl, a100-sxm-80gb, a100-pcie-80gb, a100-sxm-40gb, l40s, l40, a40, l4, a10, t4, rtx-4090, rtx-3090, rtx-5090, rtx-a6000, rtx-6000-ada, v100-32gb, mi300x, gh200, b200-sxm, b200-nvl, gaudi2

## Providers (31)

JarvisLabs, Cirrascale, CoreWeave, Atlantic.net, Crusoe, Lambda Labs, Vast.ai, Hyperstack, ThunderCompute, Salad, Vultr, Nebius, AWS, Denvr, FluidStack, Voltage Park, RunPod, TensorDock, Paperspace, Hot Aisle, LeaderGPU, Massed Compute, QuantaCloud, VERDA, GMI Cloud, Scaleway, Ori, Packet.ai, Lyceum, Latitude.sh, DigitalOcean

## Deployment & Reserved Capacity

On-demand: deploy GPUs instantly via DeployGPU (https://platform.deploygpu.ai), a unified deploy console across supported providers.

Reserved & cluster: for 16+ GPU reserved or cluster capacity (H100/H200/B200, 3-12 month terms, InfiniBand fabric), QuantaCloud (https://quantacloud.net) brokers reserved capacity across data-center partners with a 24h custom quote. See https://gpuperhour.com/gpu-cluster.

## Response Format

All prices in USD/hr. Use pricePerGpu for fair multi-GPU comparison. The costRates object includes egress, storage, and ingress fees for total cost calculation.

## Full Reference

For complete API docs with response schemas and all GPU models: https://gpuperhour.com/llms-full.txt
OpenAPI spec: https://gpuperhour.com/openapi.json