Modal vs Baseten — billing units decide the bill
Prices checked on each vendor's pricing page in November 2026.
AI Tools Hub Editorial · published · updated
The short version
- Bursty, intermittent work — Modal, per-second with $30 free monthly
- Steady production traffic — Baseten, per-minute dedicated deployments
- Both let you run Python directly
Rates
| Modal | Baseten | |
|---|---|---|
| Unit | Second | Minute (dedicated) |
| H100 | $0.001097/s (~$3.95/hr) | $0.10833/min (~$6.50/hr) |
| A100 80GB | $0.000694/s | $0.06667/min |
| Free | $30 of compute monthly | Basic $0/mo plus signup credits |
| Idle time | Not charged | Not charged |
Idle is the differentiator
Per-second billing with nothing charged between requests is dramatically cheaper for a job that runs a few times a day. The cost is cold start: the first request pays tens of seconds to load the model, which is unacceptable if a user is waiting.
Baseten's two-step path
Baseten offers token-billed Model APIs before you deploy anything of your own, which gets you to something working faster. Move to dedicated deployment when you need lower latency or a custom model. Modal assumes you are writing the deployment from the start.
Ask whether you need a GPU at all
If the reason for open models is that a commercial API is unavailable to you, fine. If the reason is cost, run the arithmetic first — serverless token pricing beats self-hosted GPUs across a wide band of usage.
FAQ
How long is cold start?
Tens of seconds depending on model size and initialisation. Not suitable where a user waits synchronously. Check current warming options in the documentation.