GPU Cloud Cost Comparison
Enter busy GPU-hours per month, pod utilization and an optional serverless cold-start overhead. The table prices H100, H200, B200, A100 80/40GB, L40S, L4, RTX 4090, A10 and T4 on Modal (per-second, scales to zero), RunPod Community and Secure pods, RunPod Serverless and Lambda 1× on-demand, and shows the utilization where an always-on pod beats Modal. Estimates only; provider pricing pages prevail.
Math runs in your browser. Nothing is uploaded.
| GPU | Modal (per-second) | RunPod Community pod | RunPod Secure pod | RunPod Serverless | Lambda 1× on-demand | Break-even util vs Community pod | Cheapest |
|---|
—
Formula: serverless (Modal, RunPod Serverless) = busy hours × (1 + overhead) × $/h — Modal $/h = $/s × 3600, scales to zero. Pods and Lambda instances bill wall-clock: busy hours ÷ utilization × $/h. Break-even utilization = RunPod Community $/h ÷ Modal effective $/h: above it an always-on pod is cheaper, below it per-second serverless wins (spiky traffic). GPU-only list prices; Modal CPU/memory, storage and egress extra. Lambda shows 1× single-GPU on-demand rows only (A100 80GB is offered only as 8× instances, so not shown). Lambda 1× also lists GH200 $2.29, H100 PCIe $3.29, A6000 $1.09, Quadro RTX 6000 $0.69.
Estimates only. Official list prices as of 2026-11-01 (Asia/Shanghai); the provider's pricing page prevails. Sources: Modal pricing · RunPod pricing · Lambda pricing
Modal GPU Cost Calculator · RunPod GPU Cost Calculator · fal.ai Cost Calculator · Together AI Token Cost Calculator · Fireworks Token Cost Calculator · DeepInfra Token Cost Calculator · Cloudflare Workers AI Cost Calculator
More in this group
- Modal GPU Cost Calculator Modal per-second GPU + CPU + memory → $/request & month.
- RunPod GPU Cost Calculator RunPod pod $/h × hours ÷ utilization vs serverless.
- fal.ai Cost Calculator fal image $/MP or $/image → per kept image & month.
- Together AI Token Cost Calculator Together serverless tokens + optional US-only.
- Fireworks Token Cost Calculator Fireworks in/cached/out + Standard/Priority/Batch.
- DeepInfra Token Cost Calculator DeepInfra tokens × Standard/Priority/Flex.
- Cloudflare Workers AI Cost Calculator Workers AI tokens + Neurons footnote.
How it works
- Which GPU cloud is cheapest for an H100?
- Per wall-clock hour RunPod Community ($2.69) is cheapest, then RunPod Secure ($3.99), Modal ($3.95 effective), Lambda 1× ($4.29) and RunPod Serverless ($5.39). But Modal and RunPod Serverless bill only busy seconds, so at low utilization they cost far less per month.
- What is the break-even utilization?
- Pod $/h ÷ serverless $/h. H100: $2.69 ÷ $3.95 ≈ 68% — if your pod would be busy less than ~68% of the time, Modal's per-second billing is cheaper; above that an always-on RunPod Community pod wins.
- Why isn't Lambda's A100 80GB shown?
- Lambda lists A100 80GB SXM ($2.79/GPU) only as an 8× instance. This comparison uses 1× single-GPU on-demand rows only: B200 $6.99, H100 SXM $4.29, H100 PCIe $3.29, GH200 $2.29, A100 40GB $1.99, A10 $1.29, A6000 $1.09.
- What is the cold-start overhead input?
- Serverless platforms bill some non-request time: container boot, model load and keep-alive after the last request. Add it as a percentage on top of busy hours for Modal and RunPod Serverless.
- Should I use a GPU API instead of self-hosting?
- For standard open models, per-token APIs (Together, Fireworks, DeepInfra, Workers AI) or fal.ai often beat renting GPUs at low volume. Compare with the linked API cost calculators.