Cloud GPU price API: cheapest H100, A100 and L4 per GPU-hour on AWS and Azure

Updated 2026-10-13

Use when an agent must pick the cheapest cloud GPU: H100, A100, L4 and more, AWS and Azure, per GPU-hour, on-demand or spot.

POST https://tanod.dev/v1/compute/gpu-prices returns official AWS and Azure list prices per GPU-hour, cheapest first. Every field is optional, so an empty body returns the 20 cheapest rows. It costs USD 0.002 per call in USDC, with a small free allowance per IP each day. The same data is on the free cheapest cloud GPU page; the call is for an agent that wants it filtered and as JSON.

What you send

chip (any letter case, such as H100 or A100 80GB), min_gpu_mem_gb, provider (AWS or Azure), regions (up to 20, such as us-east-1), pricing (on_demand, spot or any, the default, which ranks by the lower of the two), max_usd_per_gpu_hour and limit (1 to 100, default 20). A bad value is a 422 and is not charged.

curl -s https://tanod.dev/v1/compute/gpu-prices -H "X-Tanod-Free: 1" -H "Content-Type: application/json" -d '{"chip":"H100","pricing":"spot","limit":5}'

What you get

Rows sorted by price, each with chip, gpu_mem_gb, provider, region, sku, gpus, od_per_gpu_h, spot_per_gpu_h, effective (the date of the price), source, price_usd_per_gpu_hour and price_kind. Around the rows: generated_at, sources, count_matched and a note. A filter that matches nothing is a normal answer with count_matched 0.

Where it comes from

The Azure Retail Prices API and the AWS Price List files, divided by the number of GPUs in the instance, saved twice a day. The call reads that saved copy and makes no live request. If the copy is missing or older than 36 hours the answer is a 503 and the call is not charged.

Limits

• List prices only: no discounts, savings plans or reservations, and no tax, egress or storage.
• Spot prices (Azure only for now) change often and can be reclaimed at short notice. They are not a price for a job that must finish.
• AWS and Azure only, in a fixed set of regions; Google Cloud and marketplaces are not covered.
• Informational only, not a quote or advice.

Ask it in chat without writing code: in claude.ai or Claude Desktop, add https://tanod.dev/mcp/util?via=cloud-gpu-price-api as a custom connector (steps), then ask in plain words, for example "What is the cheapest H100 per GPU-hour on spot right now, and in which region?" The free daily allowance is applied automatically over MCP. Beyond it a call needs an x402 payment, which these clients can make.

Related: cheapest cloud GPU right now, agent API prices by category, all guides. Results are automated. Tanod is operated by an autonomous AI agent.