Modal
Modal pricing charges by the second for three resources at once: CPU cores, memory and GPU each carry their own rate, and a single function's bill is the sum.
Updated on:
Modal pricing: three meters running on one container
Modal pricing charges by the second for three resources at once: CPU cores, memory and GPU each carry their own rate, and a single function's bill is the sum. There's no instance type to pick and no minimum billing increment. Modal charges max(request, actual) for every second the container is alive. That decomposition is the whole design, and it's rare.
Key takeaways
Modal charges $0.0000131 per physical core-second ($0.0472 per core-hour) and $0.00000222 per GiB-second ($0.0080 per GiB-hour), independent of any GPU attached.
GPU rates run from $0.000164 per second for a T4 ($0.59/hour) to $0.001972 for a B300 ($7.10/hour), added on top of the CPU and memory charge.
Every container carries a minimum request of 0.125 cores and 128 MiB, and Modal bills the higher of your request or your real usage.
Startup time bills, and so does the default 60 seconds of idle time before a container scales down.
Modal pricing in 2026
Tier | Price | What's included | Metered limit
|
|---|---|---|---|
Starter | $0 + compute | $30/month credit, 3 seats, 200 deployed apps, 1 day log retention | 100 containers, 10 GPU concurrency |
Team | $250/month + compute | $100/month credit, unlimited seats, 30 day logs, environment budgets, static IP proxy | 5,000 containers, 50 GPU concurrency |
Enterprise | Custom | Volume discounts, SAML SSO, audit logs, HIPAA, private Slack support | Custom |
What Modal actually meters
Modal meters three resources on separate clocks and adds them up. CPU is billed per physical core per second, and Modal notes that one physical core equals two AWS or GCP vCPUs. Memory is billed per GiB-second. GPUs are billed per GPU-second by chip. Because the axes are independent, a job that's memory-hungry and GPU-light costs what it actually consumes rather than what the nearest instance type would have charged.
The rate you pay is max(request, actual) on CPU and memory, measured continuously, with a floor of 0.125 cores and 128 MiB per container. Disk isn't a separate line: requesting disk raises your memory request at a 20:1 ratio, so 500 GiB of disk bills as 25 GiB of memory.
Three multipliers sit on top. Pinning a container to a broad region such as us costs 1.15x base price, and a narrow region such as us-west costs 1.75x. Running CPU functions on non-preemptible capacity costs 3x on CPU and memory. Sandboxes and Notebooks get their own block on the pricing page at $0.00003942 per core-second and $0.00000667 per GiB-second, which is the 3x non-preemptible rate applied to base figures the page itself rounds, because Modal documents Sandboxes as non-preemptible unless they request a GPU. The pricing page prints those numbers without connecting them to the multiplier.
From 1 October 2026 network egress joins the meter at $0.04 per GiB, after 1 TiB included on Starter, 10 TiB on Team and 100 TiB on Enterprise.
What happens when you hit the limit
Modal auto-charges. There's no included compute beyond the monthly credit, so once credit runs out you pay list rate and get billed at the end of the cycle, plus incremental charges the first time you cross certain thresholds. Modal doesn't publish what those thresholds are.
Spend control is a first-class feature rather than an alert. A Workspace budget caps usage before credits apply. A separate spend limit caps net charges after credits, and Modal stops workloads that would push past it. If you don't set one, the default spend limit is the cycle's usage limit minus credits. Team and Enterprise workspaces can add Environment budgets, though those exclude storage and reservations, so an environment budget isn't an invoice total.
How Modal pricing has changed across all these years
Date | Milestone | Source
|
|---|---|---|
1 Oct 2026 | Network egress starts billing at $0.04 per GiB beyond the plan allowance | Vendor |
1 Sep 2026 | Network egress appears on the billing page, metered but not charged | Vendor |
12 Aug 2026 | Workspace.billing.rates() and modal billing rates let you query your own rate card | Vendor |
23 Jul 2026 | Billing summary APIs expose spend by category plus credit usage | Vendor |
23 Jun 2026 | Workspace and Environment billing report APIs added | Vendor |
12 Feb 2026 | modal billing report CLI reaches general availability for Team and Enterprise | Vendor |
20 Nov 2025 | Non-preemptible CPU capacity ships with a 3x multiplier on CPU and memory pricing | Vendor |
Customer
Sentiment Highlights
"Now we’re running the same workloads for under $100, with fast autoscaling and no lag."
botirk, Hacker News, October 2025
"Cold boot times are around 5m but if your usage periods are predictable it can work out ok. Works out at $2 an hour."
siquick, Hacker News, February 2026
Explore other providers
Leonardo AI
Image / Video Generation
Leonardo AI pricing sells tokens whose cost tracks the compute a job burns, not the images it returns.
Lovable
Developer Tool
Lovable pricing sells you a monthly bundle of credits, and everything you do in the product draws from it.
LangSmith
Data Platform
LangSmith pricing charges $39 per seat a month, then $0.005 for every trace past your allowance, then another $0.0025 if you want that trace to outlive 14 days.
How much does Modal cost per hour?
Does Modal have a free tier?
Does Modal bill for cold starts and idle time?
Do Modal credits roll over?
