Baseten vs Modal: Pricing, Cold Starts and What the Rate Card Hides

It depends on the GPU, and the published rates mislead. On headline numbers Modal looks 37-40% cheaper on H100 and A100 80GB. Add the CPU and memory Baseten bundles and the H100 gap narrows to about 19%, A100 80GB is parity, and on T4 and A10 Baseten is 9-22% cheaper. Modal’s pinning and non-preemptible multipliers push further toward Baseten, and its $250/mo Team fee is fixed cost Baseten does not charge.

Starter $0/mo with $30/mo free compute; Team $250/mo with $100/mo free compute; Enterprise custom. Compute is per second and itemised: GPU at its own rate, CPU at $0.04716 per physical core-hour, memory at $0.007992 per GiB-hour. Sandbox and Notebook compute is 3x the Function rate, and pinning and non-preemptible capacity carry multipliers.

Basic is $0/month pay-as-you-go with no platform fee, including SOC 2 Type II and HIPAA; Pro and Enterprise are on quote. GPU compute bills per minute against all-in instance SKUs bundling vCPU and RAM — $6.4998/hr for an H100 80GB. Per-token Model API rates are published; the free-credit amount is not.

Containers boot in roughly a second, and Memory Snapshots deliver 3-10x faster restores per Modal’s docs. The “100x” figure in Modal’s Series C post is not supported by its own documentation, so plan against 3-10x. GPU snapshots are Alpha and do not speed up loading weights from storage.

Not confirmed. The last valuation Baseten has published is $13B, from a $1.5B Series F on 22 June 2026. Press reports in late September 2026 describe talks around a larger round, but that is talks-stage reporting and the company has published nothing.