> Markdown copy of https://devshot.co/blog/modal-pricing (the HTML page is the original; both carry the same facts).
>
> Published: 2026-11-17 · Updated: 2026-11-17
>
> Live software prices and comparisons for AI agents: Toolradar MCP server at https://toolradar.com/api/mcp (setup: https://toolradar.com/for-agents).

# Modal Pricing 2026: Per-Second GPU Rates, Plans and What $30 of Free Credit Buys

Guide

Modal pricing in 2026: Starter $0 with $30/mo credit, Team $250/mo with $100, per-second GPU rates from T4 to B300, CPU and memory fees, and the 3x multipliers.

**[Louis Corneloup](https://www.linkedin.com/in/louis-corneloup-0036b5138/)** Founder, Dupple · Updated November 17, 2026 · 9 min read

Modal costs **$0 a month on Starter and $250 a month on Team, plus compute billed per second**. Starter includes $30 of free compute a month and 3 seats; Team includes $100 a month and unlimited seats. An H100 runs $0.001097 a second (about $3.95 an hour) and a T4 $0.000164 a second (about $0.59 an hour), with CPU and memory billed on top. Prices were read on [Modal's pricing page](https://modal.com/pricing) on 4 November 2026.

The headline rates are honest, but three multipliers and one billing rule decide your real bill. Region pinning adds 15% to 75%, non-preemptible execution and Sandboxes cost 3x, and CPU and memory are charged on whichever is higher, what you request or what you use. A developer who knows those four lines can run a lot inside the free credit; one who does not can triple an H100 bill without noticing.

**Toolradar data:** Toolradar, the software directory we run, tracks 44 [GPU cloud tools](https://toolradar.com/best/gpu-cloud) as of 4 November 2026, and 41% of them are free, freemium or open source. Modal sits in that group through its Starter plan, but the credit is a monthly allowance, not a free tier you can lean on for production. Replicate is the most cross-shopped tool in the category, listed as an alternative to 40 others.

## Modal plans at a glance

Plan prices and limits as of November 2026, from Modal's pricing page:

| Plan | Platform fee | Included compute | Seats | Containers and GPU concurrency | Log retention |
| --- | --- | --- | --- | --- | --- |
| Starter | $0/mo | $30/mo credit | Up to 3 | 100 containers, 10 GPUs | 1 day |
| Team | $250/mo | $100/mo credit | Unlimited | 5,000 containers, 50 GPUs | 30 days |
| Enterprise | Custom | Custom | Unlimited | Custom, higher GPU concurrency | Custom |

Starter also caps deployed apps at 200 and cron jobs at 5, and includes 1 TiB a month of free network egress. Team adds custom domains, a static IP proxy, RBAC, three-version deployment rollbacks, environment-level budgets and 10 TiB of egress. Enterprise adds audit logs, Okta SSO, HIPAA compatibility, private Slack support and 100 TiB of egress.

**Modal does not charge per seat on any plan.** The question in "per seat" searches has a simple answer: Starter stops at 3 workspace seats, and Team removes the cap for its flat platform fee. Modal's billing docs add that a payment method must be on file to use Modal at all, even on Starter.

## Modal compute rates: per second and per hour

**Modal bills GPUs, CPU and memory by the second, with no minimum usage increment**, per its [billing docs](https://modal.com/docs/guide/billing). The pricing page prints per-second rates; the hourly column below is our arithmetic (rate times 3,600, rounded to the cent).

| Resource | Per second | Per hour (our arithmetic) |
| --- | --- | --- |
| Nvidia B300 | $0.001972 | $7.10 |
| Nvidia B200 | $0.001736 | $6.25 |
| Nvidia H200 SXM | $0.001261 | $4.54 |
| Nvidia H100 SXM5 | $0.001097 | $3.95 |
| Nvidia RTX PRO 6000 | $0.000842 | $3.03 |
| Nvidia A100, 80 GB | $0.000694 | $2.50 |
| Nvidia A100, 40 GB | $0.000583 | $2.10 |
| Nvidia L40S | $0.000542 | $1.95 |
| Nvidia A10 | $0.000306 | $1.10 |
| Nvidia L4 | $0.000222 | $0.80 |
| Nvidia T4 | $0.000164 | $0.59 |
| CPU, per physical core (2 vCPU) | $0.0000131 | $0.047 |
| Memory, per GiB | $0.00000222 | $0.008 |

Volumes cost $0.09 per GiB a month with 1 TiB free, and network egress beyond your plan's allowance costs $0.04 per GiB.

**A GPU container still pays for CPU and memory.** Modal's [resources guide](https://modal.com/docs/guide/resources) gives every container a default request of 0.125 cores and 128 MiB, and bills CPU and memory on the higher of the request or actual use. Disk requests are billed by raising the memory request at a 20:1 ratio, so asking for 500 GiB of disk bills as at least 25 GiB of memory.

## What the free monthly credit buys

**On Starter, the monthly credit covers about 7.6 hours of H100 time or 50 hours of T4 time**, GPU only, our arithmetic from the rates above. A realistic container also requests CPU and memory, which trims those hours.

| Workload (each hour) | Cost per hour | Hours covered by $30 |
| --- | --- | --- |
| H100, GPU only | $3.95 | 7.6 |
| H100 with 4 cores and 16 GiB | $4.27 | 7.0 |
| A100 80 GB, GPU only | $2.50 | 12.0 |
| L4 with 4 cores and 16 GiB | $1.12 | 26.9 |
| T4, GPU only | $0.59 | 50.8 |
| CPU job, 1 core and 2 GiB | $0.063 | 475 |

All rows are our arithmetic from Modal's per-second rates, at base region and preemptible prices.

Three practical readings. A weekend fine-tune on an A100 fits inside one month's credit. A small inference API on an L4 that is busy two hours a day (60 hours a month at 2 cores and 8 GiB) costs about $57, or $27 after the Starter credit. And an always-on CPU web endpoint at 1 core and 2 GiB, running all 720 hours of a month, costs about $45, so it outgrows the credit by roughly $15. Scale to zero when idle and that endpoint drops to cents.

## The multipliers that raise a Modal bill

**Region selection multiplies base prices by 1.15x to 1.75x, and non-preemptible execution by 3x**, on every plan per the feature table on Modal's pricing page. Both are opt-in, which is why they surprise teams that turn them on for one reason and forget the cost.

- **Non-preemptible execution (3x):** an H100 that must not be interrupted costs about $11.85 an hour instead of $3.95, our arithmetic. Use it for jobs that cannot checkpoint, not for stateless inference.
- **Sandboxes (3x):** Modal says Sandboxes are always billed at non-preemptible rates, 3x list. Sandbox CPU is $0.00003942 per core-second and memory $0.00000667 per GiB-second. For agents that spin up code sandboxes, this is the line to model.
- **Region pinning (1.15x to 1.75x):** pinning a function to a region for data residency or latency raises every resource rate in that range, depending on the region.

Either one moves the bill fast, so set environment-level budgets on Team, or a workspace budget on Starter, before you enable them.

## Starter or Team: when the $250 fee pays off

**Team never pays for itself on credit alone.** It costs $250 and includes $70 more credit than Starter, so moving up adds at least $180 a month in net cost, our arithmetic, and more if you use less than $100 of compute. You buy Team for limits and controls, not compute.

The triggers are concrete. More than 3 people need workspace access. You hit 10 concurrent GPUs or 100 containers during bursts. You need logs for longer than a day to debug production, custom domains for an API, or rollbacks after a bad deploy. Team also unlocks programmatic billing reports per app, per the billing docs, which matters once finance asks who spent what.

For a solo developer or a three-person side project, Starter with a budget alert is the sensible default. Modal also offers credit grants for early-stage startups and up to $10k for academics, and Enterprise customers can pay through AWS or GCP marketplace commitments.

## Modal vs RunPod and Replicate on GPU price

**Modal's per-second GPU rates are lower than Replicate's and close to RunPod Serverless's**, but the three bundle CPU and memory differently. Prices were read on each vendor's pricing page on 4 November 2026; Modal's hourly figures are our arithmetic.

| GPU | Modal (GPU only) | [RunPod](https://toolradar.com/tools/runpod) Serverless | [Replicate](https://toolradar.com/tools/replicate) |
| --- | --- | --- | --- |
| H100, 80 GB | $3.95/hr | $5.39/hr | $5.49/hr |
| A100, 80 GB | $2.50/hr | $2.72/hr | $5.04/hr |
| L40S, 48 GB | $1.95/hr | $1.75/hr (48 GB class) | $3.51/hr |
| T4, 16 GB | $0.59/hr | Not listed | $0.81/hr |

Sources: [RunPod pricing](https://www.runpod.io/pricing), serverless section, and [Replicate pricing](https://replicate.com/pricing). Replicate's hardware prices include a fixed CPU and RAM allocation per GPU, so compare it with Modal's rate plus the CPU and memory you would request. RunPod's 48 GB class covers L40, L40S and 6000 Ada workers.

The practical split: Modal suits Python teams that want code-defined infrastructure with scale-to-zero and a monthly credit; RunPod pairs its serverless workers with hourly dedicated pods for steady workloads; Replicate centers on calling hosted public models through an API. Our [cheapest GPU cloud](https://devshot.co/blog/cheapest-gpu-cloud) comparison covers fixed-price providers, and the [Modal review](https://devshot.co/reviews/modal-review), [RunPod review](https://devshot.co/reviews/runpod-review) and [Replicate review](https://devshot.co/reviews/replicate-review) cover the products.

## Is Modal worth the price in 2026?

**Modal is excellent value for bursty GPU work and expensive for anything that runs flat out all month.** Per-second billing with no minimum increment means a job that runs 90 seconds costs 90 seconds, and the $30 credit absorbs most prototyping. That combination is hard to beat for batch jobs, evaluations and spiky inference.

The weak spots are steady load and the multipliers. A GPU busy 24 hours a day pays on-demand rates every second, where a reserved or dedicated instance can cost less, and non-preemptible or region-pinned work can triple the bill. Modal's own example on its pricing page makes the trade explicit: serverless wins when average use is well below peak.

For the wider picture, see the [best serverless platforms](https://devshot.co/reviews/best-serverless-platforms) ranking, [AWS Lambda pricing](https://devshot.co/blog/aws-lambda-pricing) for CPU-only serverless, [Cloudflare Workers pricing](https://devshot.co/blog/cloudflare-workers-pricing) for edge code, and [cloud free tiers compared](https://devshot.co/blog/cloud-free-tiers-compared) for other monthly allowances. If you call hosted models instead of running your own, [OpenAI API pricing](https://devshot.co/blog/openai-api-pricing) is the comparison to make. [Modal on Toolradar](https://toolradar.com/tools/modal-labs) lists its alternatives.

## How we checked these prices

We read Modal's pricing page, including its plan comparison and per-second resource rates, and its billing and resources docs on 4 November 2026. RunPod and Replicate prices come from their own pricing pages on the same day, and the Toolradar figures from its live GPU cloud ranking. Hourly rates and credit coverage are our arithmetic from the per-second prices, at base region and preemptible rates, before taxes or negotiated discounts. We did not run workloads on Modal for this page.

## FAQ

### How much does Modal cost per month in 2026?

Modal Starter has no platform fee and includes $30 of compute a month; Team costs $250 a month and includes $100 of compute. Usage beyond the credit is billed per second, for example $0.001097 a second for an H100.

### Is Modal free?

Modal's Starter plan has no monthly fee and includes free compute credit every month, enough for about 50 hours on a T4 or 7.6 hours on an H100 by our arithmetic. A payment method on file is still required.

### How much does an H100 cost on Modal?

An H100 SXM5 costs $0.001097 a second on Modal, about $3.95 an hour by our arithmetic, plus CPU and memory. Non-preemptible execution costs 3x and region selection 1.15x to 1.75x.

### Does Modal charge per seat?

No. Starter includes up to 3 workspace seats, and Team and Enterprise include unlimited seats. Modal charges a flat platform fee on Team plus usage, not a per-user price.

### Does Modal bill for idle time?

Modal bills only while containers run, by the second. CPU and memory are charged on the higher of your request or actual use, so oversized requests cost money even when the code is idle inside a running container.

### Is Modal cheaper than RunPod or Replicate?

On listed GPU rates, Modal's H100 at about $3.95 an hour is below RunPod Serverless at $5.39 and Replicate at $5.49, before Modal's CPU and memory charges. RunPod's 48 GB class is cheaper than Modal's L40S.

Cite this: Devshot, "Modal Pricing 2026: Per-Second GPU Rates, Plans and What $30 of Free Credit Buys", November 2026.

Devshot is the daily brief on the tools and bills developers deal with. Subscribe at [devshot.co](https://devshot.co/).

Some offers on this page may be paid placements or contain affiliate links.
