TL;DR: Among the providers Lantern Works tracks, the cheapest listed on-demand rates on Oct 27, 2026 were $1.99/hr for an H100 80 GB (RunPod Community Cloud), $1.09/hr for an A100 80 GB (Thunder Compute) and $0.34/hr for an RTX 4090 (RunPod Community Cloud). The data is open JSON, and there's a read-only MCP server so an agent can query it: uvx lantern-works-deals-mcp (MIT, no API key).
"Cheapest" in this post always means cheapest among tracked providers, not the whole market.
The problem
Comparing GPU rentals means juggling hourly vs. per-second vs. monthly billing, per-GPU vs. per-server pricing, community vs. secure tiers, and marketplace prices that move by the minute, across many tabs.
Cheapest tracked rate per GPU (checked Oct 27, 2026)
50 rates from 7 providers: 99Stack Cloud, Lambda, Modal, RunPod, Thunder Compute, Ucartz and Vast.ai. Prices are per GPU, on-demand, from each vendor's own pricing page.
| GPU | Cheapest tracked rate | Provider (tier) | Tracked rates |
|---|---|---|---|
| H100 80 GB | $1.99/hr | RunPod (Community Cloud, H100 PCIe) | 10 |
| H200 141 GB | $3.59/hr | RunPod (Community Cloud) | 3 |
| A100 80 GB | $1.09/hr | Thunder Compute (on-demand) | 10 |
| L40S 48 GB | $0.79/hr | RunPod (Community Cloud) | 3 |
| RTX 5090 32 GB | $0.69/hr | RunPod (Community Cloud) | 3 |
| RTX 4090 24 GB | $0.34/hr | RunPod (Community Cloud) | 4 |
| RTX A6000 48 GB | $0.33/hr | RunPod (Community Cloud) | 4 |
Each rate has the vendor quote it came from, a source URL and a check date. Live tables: GPU hub · full price comparison · H100 page.
A few things the table doesn't show:
- Held out, not ranked: TensorDock's public pricing table says it was last updated July 24, 2024, so it's excluded until re-checked. Vast.ai's A100 SXM 80 GB had only 3 qualifying marketplace offers on Oct 27, under our 5-offer minimum, so it's held too.
- Marketplace prices move: Vast.ai's RTX 4090 figure ($0.51/hr) is the median of live offers from Vast-verified hosts, not a fixed list price.
- Lowest hourly ≠ lowest cost per job: compare VRAM and speed for your workload. Totals exclude storage, egress and taxes. Confirm on the vendor page before you pay.
Price your own hours
- GPU cost calculator: pick a GPU and monthly hours to sort tracked providers by estimated monthly cost.
- Stack optimizer: enter monthly usage (pages scraped, proxy GB, emails, search queries, GPU hours) to get the cheapest suitable plan among tracked providers. Prices a vendor doesn't publish show as unknown and are never estimated.
Use it from code
curl -s https://thelanternworks.com/api/v1/gpu-prices.json | jq '.rates | sort_by(.hourly_equiv_usd) | .[0:5]'
Other feeds: /api/v1/deals.json and /api/v1/stack-pricing.json. There's also a standard-library-only price monitor that diffs snapshots for cron or GitHub Actions: https://github.com/connormatthewdouglas/agent-work-stack-deals/tree/main/examples/gpu-price-monitor
Use it from your agent
The MCP server (v0.3.2) is read-only, runs over stdio and has five tools: search_deals, compare_deals, get_deal, compare_gpu_prices, optimize_stack.
uvx lantern-works-deals-mcp
claude mcp add lantern-works-deals -- uvx lantern-works-deals-mcp
Setup for Cursor, Cline, VS Code, Claude Desktop and other clients: thelanternworks.com/install. There's also an agent skill: npx skills add connormatthewdouglas/agent-work-stack-deals.
Affiliate disclosure
Some listings are referral links, including RunPod and Thunder Compute, which hold most of the cheapest rows above. Lantern Works may earn a commission if you sign up through them; that never changes the price you pay. Referral links are flagged in the data (affiliate), and ranking is by estimated cost only, never by affiliate status.
What's missing
Tell us which providers or GPUs to add: https://github.com/connormatthewdouglas/agent-work-stack-deals/issues
Top comments (0)