Affiliate disclosure: HostFleet may earn a commission if you sign up through links on this page. That never changes the recommendation. Read the live HostFleet about page for methodology and affiliate-policy context. This is a source-backed pricing guide: published rates and billing behavior come from Modal’s public pricing page and documentation. The 30-day figures are clearly labeled estimates, not quotes.

Pricing verified: July 25, 2026

Modal pricing 2026: per-second GPU billing, explained

Modal does not sell a simple GPU instance with one obvious monthly price. It sells serverless compute billed by the second. That is excellent when a GPU job is genuinely intermittent; it is less magical when you keep a container warm all day to avoid a cold start.

This guide is source-backed and uses Modal’s live public pricing page checked on July 25, 2026, plus its billing, cold-start, scaling, GPU, and timeout documentation. It covers hosting infrastructure costs, not model-token or subscription pricing. For the broader provider market, start with Every serverless GPU host compared. For Modal’s product fit beyond price, see our Modal review for inference APIs and jobs.

The short answer

If you need…	Modal cost shape	Practical reading
A prototype or an intermittent GPU job	Often very good	Starter has a $0 base and $30/month credit; billed work can scale down to zero.
A Python API that sees uneven traffic	Usually a strong fit	You pay for the resources used or requested, without a minimum time increment.
One low-latency GPU held ready all month	Expensive by design	Warm capacity is billable capacity, even when it is waiting.
The cheapest predictable always-on server	Usually the wrong shape	Modal is elastic compute, not a bargain fixed GPU box.

The buying rule: use Modal for variable work; price warm capacity as if it were deliberately always on.

Modal’s live GPU rates

The table uses the live rates listed on Modal’s pricing page. The hourly column is a conversion of the vendor’s per-second rate (per-second rate × 3,600), rounded to cents. It is there to make comparison easier; Modal bills per second. All rates are USD per GPU and exclude CPU, memory, storage, and any workload-specific overhead.

GPU	VRAM	Published rate	Equivalent per hour	30-day always-allocated estimate
T4	16 GB	$0.000164/sec	$0.59/hr	about $425
L4	24 GB	$0.000222/sec	$0.80/hr	about $575
A10	24 GB	$0.000306/sec	$1.10/hr	about $793
L40S	48 GB	$0.000542/sec	$1.95/hr	about $1,405
A100	40 GB	$0.000583/sec	$2.10/hr	about $1,511
A100	80 GB	$0.000694/sec	$2.50/hr	about $1,799
RTX PRO 6000	96 GB	$0.000842/sec	$3.03/hr	about $2,182
H100	80 GB	$0.001097/sec	$3.95/hr	about $2,843
H200	141 GB	$0.001261/sec	$4.54/hr	about $3,269
B200	180 GB	$0.001736/sec	$6.25/hr	about $4,500
B300	288 GB	$0.001972/sec	$7.10/hr	about $5,111

Estimate assumptions: one GPU allocated continuously for 720 hours in a 30-day month; GPU charges only; no CPU, memory, volumes, networking, or region-related charges. Formula: per-second rate × 3,600 × 720. These estimates describe a continuously allocated GPU, not an intermittent endpoint that actually scales down. Re-check the Modal pricing page before committing a production workload.

Starter and Team: the fixed plan cost is not the whole bill

Modal’s current public plan table lists Starter at $0 plus compute with $30/month in free credit, three workspace seats, 100 containers, and 10 GPU concurrency. Scheduled and Web Functions are marked limited. Team is $250 plus compute with $100/month in free credit, 1,000 containers, 50 GPU concurrency, and unlimited Scheduled Functions.

The important wording is “plus compute.” The plan is a control-plane and feature gate; GPU, CPU, memory, and storage remain usage charges. Treat credits as a prototype runway, not as a production cost model.

What actually bills

Modal’s billing guide says customers pay for resources they use or request, with no reservations required and no minimum usage-time increments. The live pricing page also lists these supporting resources:

Resource	Published price	Why it matters
CPU	$0.0000131 per physical core-second	A small API wrapper still uses CPU while it handles requests or stays allocated.
Memory	$0.00000222 per GiB-second	A large model-serving process can make RAM material alongside GPU time.
Volumes	$0.09 per GiB-month	Persistent model data and artifacts are separate from GPU charges.

The CPU rate has a documented minimum allocation of 0.125 cores per container. The pricing page also includes 1 TiB/month of volume transfer. Those details are small compared with a warm H100, but they are why a “GPU-only” estimate is not a total bill.

The warm-capacity tradeoff, in plain language

A cold container has to start before it can serve work. Modal’s cold-start guide says containers stay idle for a short period before shutdown; the default maximum idle time is 60 seconds. Its scaling controls let you use min_containers to hold capacity warm, buffer_containers to keep an extra idle buffer while active, and scaledown_window to change how long containers may remain idle.

Those controls improve latency, but Modal is explicit about the trade: a larger warm pool or longer idle window increases resource consumption and cost. A retained GPU is still a billed GPU.

For example, the L4 rate is $0.000222/second. If the deployment only does 100 billable GPU-hours of work in a month, the GPU component is about $80. If an L4 remains allocated for all 720 hours, it is about $575 before CPU, memory, and storage. The same distinction is why a permanently warm H100 is a roughly $2,843/month GPU-only decision, not an inexpensive serverless default.

Limits that should change the architecture

Modal’s default Function execution timeout is 300 seconds. The documented range is one second through 24 hours. That gives batch and async jobs room, but it is not a reason to make every long operation a synchronous web request.

A practical deployment split is:

• keep short inference calls on the request path;
• send longer document processing, fine-tuning preparation, or bulk work to an async job; and
• expose polling, a callback, or a queue status rather than holding a user connection open.

This preserves the real advantage of per-second serverless billing: work can finish and scale down instead of becoming accidental always-on capacity.

Which Modal price should you use?

Start by selecting VRAM and then match the workload’s traffic shape:

• T4 or L4: small inference, embeddings, OCR, and bursty jobs where a 16–24 GB class is enough.
• L40S or A100: larger serving workloads and models that need more headroom, with a meaningful jump in the cost of keeping containers warm.
• H100, H200, B200, or B300: high-throughput or large-memory work where utilization must be deliberate; idle capacity becomes a multi-thousand-dollar monthly decision.

If traffic is steady enough that a GPU should remain allocated for most of the month, compare the warm Modal estimate with dedicated infrastructure rather than assuming serverless is automatically cheaper. The practical tradeoff is covered in Modal vs RunPod for one warm inference endpoint and RunPod, Modal, and Replicate for a small inference API.

Final verdict

Modal pricing is straightforward once you separate billable work from billable readiness:

• Starter is a strong low-friction way to prototype intermittent GPU jobs.
• Per-second billing is valuable when the deployment truly scales down.
• Warm-container settings are latency controls that deliberately raise the bill.
• Always calculate a 720-hour GPU-only floor before calling a public endpoint “serverless and cheap.”

For an application that needs a deliberately always-on GPU rather than elasticity, a RunPod Pod can be a cleaner operational and cost model. For CPU-first agent backends that call external model APIs, renting any GPU may be unnecessary; Best hosting for AI agents on a budget explains that boundary.

Sources

• Modal pricing — checked July 25, 2026
• Modal billing guide — checked July 25, 2026
• Modal GPU guide — checked July 25, 2026
• Modal cold-start guide — checked July 25, 2026
• Modal scaling guide — checked July 25, 2026
• Modal timeouts guide — checked July 25, 2026
• HostFleet GPU pricing dataset — /opt/hostbot-v2/src/data/gpu-pricing.json, refreshed July 23, 2026

Signing up for something covered here? If your use case calls for a deliberately always-on RunPod Pod, using our affiliate link supports HostFleet’s testing budget at no extra cost: RunPod (+$5 credit on your first $10). Links are labeled, and source citations in this article are never affiliate links.