GPU batch on Modal architecture
Bursty GPU work triggered from a queue, with inputs and outputs in cheap object storage.
Every resource, and what it costs.
Projections from August 2026 list prices for always-on resources. Connect an account and these become the figures your provider actually bills.
| Node | Type | What it is | Projected |
|---|---|---|---|
| Submit API | Compute | 1 GB RAM · light vCPU | from $12/mo |
| Job queue | Database | Fixed 250 MB plan | from $10/mo |
| GPU workers | Compute | ~50 h A10G at $1.10/h | from $55/mo |
| Inputs | Storage | 500 GB · no egress fee | $8.00/mo |
| Outputs | Storage | 500 GB · no egress fee | $8.00/mo |
Agent steps are priced from provider-reported token usage once the agent runs, not estimated. Guardrails cost nothing and are the reason a runaway agent cannot. 3 rows show from because those services bill by usage, so the total is a scenario at the stated volumes, not a quote.
Similar templates.
Vercel + Supabase app
The default modern startup stack: hosted front end, managed Postgres and auth, edge cache, payments.
Fly.io app and database
Machines close to users, Postgres in the same region, Redis for sessions and rate limits.
Railway monolith and worker
One web service, one background worker, a serverless Postgres and object storage. The smallest real stack.
Open GPU batch on Modal on the canvas.
It loads as an editable graph. Connect an account or instrument an agent and the projected figures above become measured ones.