Image: Unknown · CC0 1.0 Public Domain Dedication · Openverse
Serverless containers for Python workloads with per-second GPU billing and no charge for idle capacity.
What it changes
Lets a team run training, inference and batch jobs on GPUs without reserving machines, so spend tracks compute used rather than capacity parked.
Availability
Generally available (global)
Access and cost
Get Modal — Per-second billing with published GPU rates, from $0.000164 per second on a T4 to $0.001972 on a B300, plus a free tier.
Evidence it works
- Publishes per-second GPU pricing, with H100 SXM5 at $0.001097 per second and A100 40GB at $0.000583. — source
- The free Starter plan includes $30 of monthly compute and three workspace seats. — source
What it cannot do yet
- Designed to scale to zero, so it suits bursty jobs better than always-on low-latency serving.
- Storage volumes bill monthly at $0.09 per GiB after the first free terabyte.
Modal Labs · Infrastructure & Compute · Last verified 2026-09-17. Claims are re-checked on a 30-day clock; the date above is when this line was last confirmed.
About the Author
