Modal review: an inference provider for running your own code on serverless GPUs
Answer in brief: Modal ranks #9 among the best AI inference providers. It is the pick for engineers who want their own serving stack without managing clusters. You write Python, deploy LLM, image, video or audio inference, and pay per second with scale to zero. An H100 is $0.001097/sec (about $3.95/hr), and the free Starter plan includes $30/month in credits. You build and operate the serving stack yourself.
Independent research, checked 2026-11-09 against Modal's pricing page. We did not test the product hands-on.
Your code, their GPUs, billed by the second
Modal is a serverless compute platform rather than a token API. You bring the model and the serving code; Modal handles containers, autoscaling and GPUs. It bills during application load time and while containers process inputs. By default containers stay alive for 60 seconds after the last input, which is also billed and configurable. Once you scale to zero you stop paying. Modal says containers spin up and down in 1–2 seconds, which is why it argues that spiky workloads cost less than on reserved GPUs.
Who Modal fits best
Modal fits ML engineers running custom or fine-tuned models, teams with spiky or unpredictable traffic, and builders who want one platform for inference, batch jobs, training and Sandboxes.
Why teams pick Modal
- Per-second GPU pricing: H100 $0.001097/sec, H200 $0.001261/sec, B200 $0.001736/sec, A100 80GB $0.000694/sec, L4 $0.000222/sec.
- Scale to zero: no charge once containers have shut down.
- Generous starter credits: $30/month on Starter and $100/month on Team, plus credit programs for startups and up to $10k for academics.
- Security: SOC 2 on every plan; HIPAA, audit logs and Okta SSO on Enterprise.
- Procurement: Enterprise customers can use AWS and GCP marketplace committed spend on Modal.
The honest drawbacks
- You own the serving stack: batching, quantization and model optimization are on you.
- Region selection costs 1.15–1.75x base prices, and non-preemptible execution costs 3x.
- GPU concurrency is capped at 10 on Starter and 50 on Team.
- The default 60-second keep-alive after the last input is billed.
Pricing at a glance
| Item | Published price |
|---|---|
| Nvidia H100 | $0.001097/sec (about $3.95/hr) |
| Nvidia B200 | $0.001736/sec |
| Nvidia B300 | $0.001972/sec |
| Starter plan | $0 + compute; $30/month credits; 10 GPU concurrency |
| Team plan | $250/month + compute; $100/month credits; 50 GPU concurrency |
Best Modal alternatives
- Baseten: managed dedicated deployments plus cheap Model APIs.
- Replicate: deploy your own models with Cog, or call thousands of community models.
- Together AI: managed serverless tokens, dedicated endpoints and GPU clusters in one place.
Modal FAQ
What counts as billable time on Modal?
Application load time, time spent processing inputs, and the container keep-alive window (60 seconds by default) after the last input.
Do I need to buy credits up front on Modal?
No. Starter and Team are pay-as-you-go, with included monthly credits, and you can set a workspace budget.
Is Modal HIPAA compliant?
HIPAA compatibility is included on the Enterprise plan; SOC 2 applies to all plans.
Sources
- Modal pricing
← Back to the ranked list of the best AI inference providers
Named on this page? Join the network for a linked profile, Visit buttons and featured placement.