Modal GPU pricing: Modal rates, September 2026
Modal lists H100 SXM5 at about US$3.95 per GPU-hour. Compare the full workload bill: CPU, memory, plan fees and execution options are separate cost inputs.
Modal lists the H100 SXM5 at $0.001097 per GPU-second, about US$3.95 per GPU-hour, on its pricing page, checked 28 September 2026.1 Its listed GPUs range from about US$0.59 per hour for a T4 to US$7.10 for a B300.1 CPU, memory and any monthly plan fee are separate charges.1
TL;DR
- The hourly GPU figures are equivalents of Modal’s per-second rates. Use them to compare prices, then estimate spending from billable GPU time.1
- Starter has no monthly plan fee and includes US$30 of compute per month. Team costs US$250 per month plus compute and includes US$100 of compute.1
- Sandbox and Notebooks use higher CPU and memory rates than standard compute; their GPU pricing points to the standard GPU tariff.1
- Modal lists region selection at 1.15 - 1.75x base prices and non-preemptible execution at 3x base prices.1
How were Modal’s hourly rates calculated?
Modal’s hourly GPU equivalents are its published per-second rates multiplied by 3,600, using the pricing page checked 28 September 2026.1 GPU-hour figures are rounded to two decimal places; CPU and memory conversions retain their full decimal values.1
The comparison retains Modal’s quoted dollar figures without currency conversion or added tax. Its original parent table uses a USD, before-tax convention, but the captured pricing page does not establish those billing terms; treat them as assumptions and confirm them for your account. GPU rows retain Modal’s order and names, with both A100 memory sizes kept separate.1 Standard compute and Sandbox/Notebooks CPU and memory tariffs also remain separate.1 The research methodology explains the evidence standard.
What does each Modal GPU cost per second and per hour?
Modal’s listed GPU rates run from $0.000164 per second for the T4 to $0.001972 per second for the B300, equivalent to about US$0.59-US$7.10 per GPU-hour.1
Base list rates, checked 28 September 2026; quoted dollar figures unchanged. Hourly figures are calculated equivalents, rounded to cents.1
| GPU | USD per GPU-second | Approx. USD per GPU-hour |
|---|---|---|
| Nvidia B300 | $0.001972 | $7.101 |
| Nvidia B200 | $0.001736 | $6.251 |
| Nvidia H200 SXM | $0.001261 | $4.541 |
| Nvidia H100 SXM5 | $0.001097 | $3.951 |
| Nvidia RTX PRO 6000 | $0.000842 | $3.031 |
| Nvidia A100, 80 GB | $0.000694 | $2.501 |
| Nvidia A100, 40 GB | $0.000583 | $2.101 |
| Nvidia L40S | $0.000542 | $1.951 |
| Nvidia A10 | $0.000306 | $1.101 |
| Nvidia L4 | $0.000222 | $0.801 |
| Nvidia T4 | $0.000164 | $0.591 |
For an H100 SXM5, the unrounded conversion is $0.001097 × 3,600 = US$3.9492 per GPU-hour.1 The US$3.95 figure describes the GPU charge alone; CPU and memory have their own rates.1
Explore the compute infrastructure research for related cost questions.
What do Modal’s plans cost?
Modal’s Starter plan has a US$0 monthly fee with US$30 of included compute; Team has a US$250 monthly fee with US$100 of included compute; Enterprise pricing and included compute are custom.1
Plan terms checked 28 September 2026; quoted dollar figures unchanged.1
| Plan | Monthly plan fee, plus compute | Included compute per month | Workspace seats | GPU concurrency |
|---|---|---|---|---|
| Starter | $0 | $30 | Up to 3 | 101 |
| Team | $250 | $100 | Unlimited | 501 |
| Enterprise | Custom | Custom | Unlimited | Custom1 |
Treat the Team fee and compute allowance as separate entries in a budget: the listed US$100 allowance applies to compute alongside the US$250 plan fee.1 Starter and Team also differ in scale limits: Modal lists 100 containers for Starter and 5,000 for Team.1 Enterprise lists volume-based discounts with custom pricing.1
How much do CPU and memory cost on Modal?
Modal lists standard CPU at $0.0000131 per physical core-second and memory at $0.00000222 per GiB-second; Sandbox and Notebooks have separate, higher rates.1 Those rates belong alongside GPU time in a workload estimate.1
Resource rates checked 28 September 2026; quoted dollar figures unchanged.1
| Compute tariff | Resource unit | USD per unit-second | Calculated USD per unit-hour |
|---|---|---|---|
| Standard | Physical CPU core | $0.0000131 | $0.047161 |
| Standard | GiB of memory | $0.00000222 | $0.0079921 |
| Sandbox + Notebooks | Physical CPU core | $0.00003942 | $0.1419121 |
| Sandbox + Notebooks | GiB of memory | $0.00000667 | $0.0240121 |
Modal defines a physical core as equivalent to two vCPUs and lists a minimum of 0.125 cores per container for both tariffs.1 Keep the physical-core unit when comparing CPU prices. The listed memory unit is GiB, so preserve that unit too.1
The Sandbox and Notebooks GPU entry refers back to standard GPU pricing.1 Use the Sandbox and Notebooks tariff for CPU and memory alongside the standard GPU rate.
Which charges can raise the base price?
Modal lists region selection at 1.15 - 1.75x base prices and non-preemptible execution at 3x base prices across its plans.1 Check the execution options before using a base GPU rate in a budget.
Storage has a separate billing unit: Modal lists Volumes at US$0.09 per GiB per month, with 1 TiB per month included free.1 A monthly storage rate belongs in a separate budget line from hourly compute.
For alternatives, compare H100 rental prices by provider or B200 rental prices by provider. Keep the GPU variant, billable duration and additional resource charges consistent in the comparison.
What do hourly GPU rates leave out?
Glamdring Research’s conclusion is to compare the whole workload bill: Modal’s GPU-hour equivalent leaves CPU, memory, storage and plan fees to be added separately.1 The listed execution multipliers can also change the base-price estimate.1
Modal’s billing FAQ, checked 1 October 2026, says billable container time includes application loading, input processing and a configurable keep-alive period after processing.1 It also says CPU and memory charges use whichever is higher: the amount requested or the amount consumed.1 Use billable container time and metered resources in the estimate, including time outside the main GPU calculation.
A GPU price table alone cannot establish cost per completed job when CPU, memory and other charges are separate inputs.1 Compare a representative workload’s duration and resource use before drawing a total-cost conclusion.
Frequently asked questions
How much does A100 cost per hour?
On Modal’s pricing page, checked 28 September 2026, the A100 40 GB is $0.000583 per second, about US$2.10 per GPU-hour.1 The A100 80 GB is $0.000694 per second, about US$2.50 per GPU-hour.1 Use the rate for the requested memory size and account for CPU and memory charges separately.1
Do I need to buy credits before using Modal?
No.1 Modal’s pricing FAQ, checked 1 October 2026, says Starter and Team usage is pay-as-you-go and does not require pre-purchasing compute or tokens.1 The monthly included compute is an allowance; paid usage can still accrue under those plans.1
Sources checked
Company-owned pages establish what a company says. They do not prove a market conclusion. Each source is dated so readers can judge each claim.
