Modal GPU Pricing (2026): Per-Second Rates for H100, A100, T4
Modal.com GPU pricing verified July 2026: H100 $3.95/h, A100 80GB $2.50/h, T4 $0.59/h, billed per second. Plus the CPU and memory charges that land on top of every invoice.
Modal (modal.com) rents GPUs by the second through a Python-first serverless runtime. There is no instance to size and no hourly minimum: you decorate a function, Modal starts a container when it is called, and billing stops when the container scales to zero. The rates below were verified against modal.com/pricing on 31 July 2026.
Modal GPU rates
| GPU | Per second | Per hour |
|---|---|---|
| B300 | $0.001972 | $7.10 |
| B200 | $0.001736 | $6.25 |
| H200 | $0.001261 | $4.54 |
| H100 | $0.001097 | $3.95 |
| A100 80GB | $0.000694 | $2.50 |
| A100 40GB | $0.000583 | $2.10 |
| L40S | $0.000542 | $1.95 |
| A10 | $0.000306 | $1.10 |
| L4 | $0.000222 | $0.80 |
| T4 | $0.000164 | $0.59 |
The CPU and memory charge nobody budgets for
The GPU rate is not the whole bill. Modal charges CPU at $0.0000131 per core-second with a floor of 0.125 cores per container, and memory at $0.00000222 per GiB-second. Both accrue alongside the GPU for as long as the container is up.
On an H100 container with 16 GiB of memory, that adds about $0.13/h for the RAM and $0.006/h for the minimum CPU slice, taking the effective rate to roughly $4.08/h rather than the advertised $3.95/h. The gap widens with memory-heavy inference: 64 GiB pushes the memory line alone past $0.51/h.
Plans
Starter has no base fee, includes $30 of credit each month, and caps you at 3 seats, 10 concurrent GPUs and 100 containers. Team costs $250/month, includes $100 of credit, and lifts those limits to unlimited seats, 50 concurrent GPUs and 5,000 containers. The $250 is a platform fee, not a prepayment — usage bills on top of it once the $100 credit is gone.
How Modal compares on the same GPU
Modal is roughly twice the price of a rented instance for identical silicon. An H100 costs $3.95/h on Modal against $1.99/h on RunPod Community Cloud; an A100 80GB is $2.50/h against $1.19/h. You are paying for scale-to-zero and the deployment model, not for faster hardware.
That premium pays for itself when utilisation is low. A model serving 90 minutes of real traffic per day costs about $5.93/day on Modal’s H100 and $47.76/day on a RunPod pod left running around the clock. Once your GPU is busy more than roughly half the day, the rented pod is cheaper. See the serverless GPU comparison for cold-start and concurrency differences.
Where Modal sits against the wider market
| Provider | Entry rate | Billing model |
|---|---|---|
| Vast.ai | $0.03/h | Marketplace, per-minute |
| RunPod | $0.16/h | Per-second pods + serverless |
| Paperspace | $0.45/h | Per-hour instances |
| AWS GPU (EC2) | $0.53/h | Per-second, 60s minimum |
| Azure GPU | $0.53/h | Per-second |
| Modal | $0.59/h (T4) | Per-second serverless |
| Lambda Labs | $0.69/h | Per-minute instances |
| Google Cloud GPU | $0.71/h | Per-second, 60s minimum |
| CoreWeave | $1.25/h | Per-node, billed hourly |
Entry rates are verified provider floors as of July 2026 and refer to different GPUs, so read them as a price band rather than a like-for-like comparison. The full GPU cloud comparison matches providers on identical hardware.
FAQ
Does Modal sell GPUs?
No. Modal does not sell or lease hardware, and there is no reserved-capacity contract to sign. You rent GPU time by the second on shared infrastructure, and the meter stops when your container scales to zero. If you need a machine you keep, a dedicated server from Hetzner or a reserved instance from a hyperscaler is the category you want.
What does Modal charge for an A100?
$2.50/h for the 80GB card ($0.000694/s) and $2.10/h for the 40GB card ($0.000583/s), before CPU and memory. Modal does not list a 48GB A100 — that capacity does not exist in NVIDIA’s A100 line, which shipped only in 40GB and 80GB variants.
Is there a free tier?
The Starter plan carries no base fee and refreshes $30 of credit monthly, which buys about 7.5 hours on an H100 or roughly 50 hours on a T4. Credit does not roll over, so unused balance is lost at the end of each month.
How does per-second billing change what I pay?
You are charged for container uptime, not request time, and that includes the cold start. A 20-second model load on an H100 costs about $0.022 every time a container spins up from zero. Bursty traffic that repeatedly triggers cold starts can add a meaningful share to the bill, which is why Modal lets you keep warm containers pinned at the usual per-second rate.