Cheapest H100 and H200 cloud in 2026: 8-hour prices
We priced one 8-hour job on an 8x H200 node, an 8x H100 node and one H100 across Lium, RunPod, Vast.ai, Lambda and AWS. Every price is dated and sourced.
Lium team6 min read
gpu-pricing
h100
h200
comparison
Eight H200s on Lium's secure tier cost $236.16 for eight hours in our 22 October price check, 15% less than the cheapest available Vast.ai node. You can also schedule pod removal and give each agent a spending budget before it starts renting GPUs.
Lium bills per second, needs no quota request or commitment, and a secure-tier provider is penalized if your rental is interrupted. Last checked: 22 Oct 2026, 23:31 UTC. Lium and Vast.ai are marketplaces where hosts set the price, so follow the source links for today's numbers before you rent.
How we priced it
We compared the cost of eight hours of GPU rental across providers:
- Job: 8 hours of on-demand compute. No reserved or committed pricing.
- GPUs: 8x H200 141GB as one node, 8x H100 80GB as one node, and 1x H100 80GB (SXM where a cloud offers it).
- Cost: the price per GPU-hour as shown, times the number of GPUs, times 8. Storage, bandwidth and tax are not included.
- Price: the lowest on-demand price a new customer could rent at that moment. On the marketplaces that is the lowest free offer.
On Lium, a secure node is one where the provider is penalized if your rental is interrupted, and a spot node can be reclaimed at any time without compensation to you. Lium, RunPod, Vast.ai and AWS bill per second (Vast.ai billing).
Eight H200s on Lium
- Lium: the $3.69 rate is for the whole node: you rent all eight GPUs together, at $29.52 per node-hour. The node is on the secure tier and was listed in the nodes feed with all eight GPUs free.
- Vast.ai: five 8x H200 offers were rentable. The cheapest, $34.74 per hour, was on an unverified host; the cheapest verified one was $35.80. Vast.ai also bills internet traffic per TB on top, set by each host.
- RunPod: the pricing page publishes $5.29 per H200 per hour on Secure Cloud but doesn't say whether eight can be rented as one pod, so the node price is that rate times eight.
- Lambda: its page lists the GH200, a different GPU, and no H200.
Need a single H200? Lium lists those too; check the live price on the H200 page.
8x H100 80GB for 8 hours
Both Lium prices are for whole 8-GPU nodes. Lium's spot nodes can be reclaimed at any time, so they suit jobs that restart from a checkpoint. Vast.ai had two 8x H100 offers, both on unverified hosts, and no verified one.
1x H100 80GB for 8 hours
- Lium: one GPU was free at $1.7945 on a secure-tier H100 node. Providers set their own prices, so
lium ls --gpu H100shows today's. - Vast.ai: five single-GPU on-demand H100 SXM offers were rentable; the cheapest was on an unverified host. Vast.ai also bills internet traffic per TB and storage on top.
- RunPod Community Cloud runs on third-party hosts, and Secure Cloud in RunPod's partner data centers.
What you get beyond the hourly price
Lium bills per second, so an 8-hour job that ends at 7 h 40 m costs 7 h 40 m. Renting needs a balance that covers 15 minutes of the pod, with no quota request and no commitment. On AWS the smallest H200 size is a whole 8-GPU instance, and a new account may need a quota increase before the first GPU launches.
A pod that fails to start isn't charged, and if the node was at fault Lium starts it on another node with the same GPU at the same or a lower price. Lium doesn't meter a pod's own traffic; only reads from a network volume are billed as egress. On Vast.ai, internet traffic is billed per TB, and on a hyperscaler moving a large dataset or checkpoint out can cost more than the GPU hours.
Renting an H100 on Lium for 8 hours
Check the live H100 price on the H100 page, no account needed.
Install the CLI and sign in once (CLI quickstart):
bash
curl -fsSL https://lium.io/install.sh | bash # or: pip install lium.io
lium initPick the node first. List the secure-tier H100 nodes at or under the price in the table, cheapest first, and check that the GPU column of the first row reads H100 80GB HBM3 (not H100 NVL).
--tier and --max-price need Lium CLI 0.0.40 or newer:bash
lium ls --gpu H100 --tier secure --max-price 1.80Rent one GPU on that node by its index in the list (row 1 here);
--count 1 rents one GPU even on a node with more free. --ttl 8h schedules the pod's removal after eight hours. At the snapshot rate of $1.7945 per GPU-hour, eight hours of compute costs approximately $14.36. Confirm the listed rate before renting; storage and other applicable charges are additional:bash
lium up 1 --count 1 --ttl 8h --name h100-job -yIf the list is empty, the $1.7945 node is taken: raise
--max-price, and the 8-hour cost rises with it.When the job finishes early, remove the pod so billing stops:
bash
lium rm h100-jobThe same works from an agent such as Claude Code or Codex: it runs these commands, reads
lium ls --format json and picks the GPU itself, on its own API key with a daily budget the server enforces. Agents can read live Lium prices without an account from lium.io/pricing.json and the nodes feed, and lium.io/llms.txt describes the agent workflow. Prices for both GPUs are also on the H100 and H200 pages and on /pricing.How we keep this updated
We re-read every price in this post from the same sources and change the tables, the totals and the "Last checked" date together. When a cloud stops publishing a price or we can't read it, the cell says so instead of keeping an old number.
Find your next GPU on Lium
Run it on Lium
Get new posts by e-mail: Subscribe to the Lium newsletter
