Modal — free tier setup

This is a compute platform, not a turnkey inference API. Modal does NOT slot into an OpenAI-client cascade: there is no hosted model catalog and no fixed completions base URL. You get monthly GPU credits and you deploy your own model server (e.g. vLLM); each deployment gets its own per-app URL. api.modal.com in the catalog is the platform API root — do not route completions there. If you want a drop-in free endpoint, use a different provider; use Modal when you want a specific open-weights model served your way, paid for by renewable credits.

What you get free

Renewable monthly compute credits, burned by GPU-seconds — not tokens. Catalog raw_text: "Credits: $5/month upon sign up, $30/month with payment method added — Models: Any supported model - pay by compute time (cheahjs). Modal pricing page: Starter plan includes "$30 / month free credits", "100 containers + 10 GPU concurrency", "5 deployed crons"; GPU billed per second (e.g. H100 "$0.001097 / sec")." Category: conditional_free — the headline $30/month has historically required adding a payment method ($5/month without, per cheahjs), though today's pricing page shows "$30 / month free credits" on Starter without stating that condition (discrepancy unconfirmed — verify on first signup). The gotcha: signup alone gets you nothing callable — you must write and deploy an inference endpoint first.

Friction check

Credit card	no for signup; possibly required to unlock the full $30/mo credit (see discrepancy above)
Phone verification	unknown
Org/membership	no (workspace created automatically)
Est. time to first key	~5 min for token; ~30–60 min to a callable model endpoint (you deploy it yourself)
Trains on your prompts	unknown — you run your own containers, so prompt handling is largely under your control; platform-level logging policy not verified (catalog)