Serverless GPU platforms compared: Modal, Runpod Serverless and Baseten
How three serverless GPU platforms bill (per second or per minute, GPU-only or bundled), what scale-to-zero costs you, and which fits which inference workload.
Choose Modal if…
- Your team writes Python and wants functions, batch jobs and endpoints defined in code, scaling to zero
- Your work tolerates preemption, because the listed GPU rates are base prices (non-preemptible execution is 3x)
- You will account for CPU and memory, which are billed separately from the GPU
Choose Runpod if…
- You already package workloads as containers and want an endpoint whose flex workers bill per second only while running
- You want one rate per GPU tier, with the worker's CPU and RAM included in it
- You may later move steady traffic onto Runpod pods or reserved active workers
Choose Baseten if…
- You are deploying a model as a production API and want a platform built around model serving
- You prefer instance types with a fixed vCPU/RAM bundle and per-minute billing
- You need multi-GPU instance types (2, 4 or 8 GPUs) behind one deployment
| Provider | GPU | Offering | USD / GPU-hour | Billing | Commitment | Regions | Source |
|---|---|---|---|---|---|---|---|
| Modal | NVIDIA B300288 GB | ServerlessServerless function (base rate) | $7.10listed: $0.001972/sec | per second | None stated | UNSPECIFIED | Modal pricing8 Nov 2026 |
| Runpod | NVIDIA B300288 GB | ServerlessServerless worker (B300) | $11.99listed: $11.99/hr | per second | None stated | UNSPECIFIED | Runpod pricing8 Nov 2026 |
| Modal | NVIDIA B200192 GB | ServerlessServerless function (base rate) | $6.25listed: $0.001736/sec | per second | None stated | UNSPECIFIED | Modal pricing8 Nov 2026 |
| Baseten | NVIDIA B200192 GB | ServerlessDedicated deployment, B200 instance | $9.98listed: $0.16633/min | per minute | None stated | UNSPECIFIED | Baseten docs: instance types8 Nov 2026 |
| Runpod | NVIDIA B200192 GB | ServerlessServerless worker (B200) | $10.79listed: $10.79/hr | per second | None stated | UNSPECIFIED | Runpod pricing8 Nov 2026 |
| Modal | NVIDIA H200141 GB | ServerlessServerless function (base rate) | $4.54listed: $0.001261/sec | per second | None stated | UNSPECIFIED | Modal pricing8 Nov 2026 |
| Runpod | NVIDIA H200141 GB | ServerlessServerless worker (H200) | $6.99listed: $6.99/hr | per second | None stated | UNSPECIFIED | Runpod pricing8 Nov 2026 |
| Baseten | NVIDIA H200141 GB | ServerlessDedicated deployment, H200 instance | $7.50listed: $0.12500/min | per minute | None stated | UNSPECIFIED | Baseten docs: instance types8 Nov 2026 |
| Modal | NVIDIA H100 SXM80 GB | ServerlessServerless function (base rate) | $3.95listed: $0.001097/sec | per second | None stated | UNSPECIFIED | Modal pricing8 Nov 2026 |
| Runpod | NVIDIA H100 (form factor not stated)80 GB | ServerlessServerless worker (H100) | $5.39listed: $5.39/hr | per second | None stated | UNSPECIFIED | Runpod pricing8 Nov 2026 |
| Baseten | NVIDIA H100 (form factor not stated)80 GB | ServerlessDedicated deployment, H100 instance | $6.50listed: $0.10833/min | per minute | None stated | UNSPECIFIED | Baseten docs: instance types8 Nov 2026 |
| Modal | NVIDIA A100 80GB80 GB | ServerlessServerless function (base rate) | $2.50listed: $0.000694/sec | per second | None stated | UNSPECIFIED | Modal pricing8 Nov 2026 |
| Runpod | NVIDIA A100 80GB80 GB | ServerlessServerless worker (A100) | $2.72listed: $2.72/hr | per second | None stated | UNSPECIFIED | Runpod pricing8 Nov 2026 |
| Baseten | NVIDIA A100 80GB80 GB | ServerlessDedicated deployment, A100:12x144 instance | $4.00listed: $0.06667/min | per minute | None stated | UNSPECIFIED | Baseten docs: instance types8 Nov 2026 |
| Modal | NVIDIA A100 40GB40 GB | ServerlessServerless function (base rate) | $2.10listed: $0.000583/sec | per second | None stated | UNSPECIFIED | Modal pricing8 Nov 2026 |
| Runpod | NVIDIA L40S48 GB | ServerlessServerless worker (L40, L40S, 6000 Ada, MIG 48GB tier) | $1.75listed: $1.75/hr | per second | None stated | UNSPECIFIED | Runpod pricing8 Nov 2026 |
| Modal | NVIDIA L40S48 GB | ServerlessServerless function (base rate) | $1.95listed: $0.000542/sec | per second | None stated | UNSPECIFIED | Modal pricing8 Nov 2026 |
| Runpod | NVIDIA L424 GB | ServerlessServerless worker (L4, A5000, 3090, MIG 24GB tier) | $0.75listed: $0.75/hr | per second | None stated | UNSPECIFIED | Runpod pricing8 Nov 2026 |
| Modal | NVIDIA L424 GB | ServerlessServerless function (base rate) | $0.799listed: $0.000222/sec | per second | None stated | UNSPECIFIED | Modal pricing8 Nov 2026 |
| Baseten | NVIDIA L424 GB | ServerlessDedicated deployment, L4:4x16 instance | $0.848listed: $0.01414/min | per minute | None stated | UNSPECIFIED | Baseten docs: instance types8 Nov 2026 |
| Modal | NVIDIA A10 / A10G24 GB | ServerlessServerless function (base rate) | $1.10listed: $0.000306/sec | per second | None stated | UNSPECIFIED | Modal pricing8 Nov 2026 |
| Baseten | NVIDIA A10 / A10G24 GB | ServerlessDedicated deployment, A10Gx4x16 instance | $1.21listed: $0.02012/min | per minute | None stated | UNSPECIFIED | Baseten docs: instance types8 Nov 2026 |
| Modal | NVIDIA RTX PRO 6000 Blackwell96 GB | ServerlessServerless function (base rate) | $3.03listed: $0.000842/sec | per second | None stated | UNSPECIFIED | Modal pricing8 Nov 2026 |
| Runpod | NVIDIA RTX PRO 6000 Blackwell96 GB | ServerlessServerless worker (RTX 6000 Pro) | $3.49listed: $3.49/hr | per second | None stated | UNSPECIFIED | Runpod pricing8 Nov 2026 |
| Baseten | NVIDIA RTX PRO 6000 Blackwell96 GB | ServerlessDedicated deployment, RTX-PRO-6000 instance | $4.00listed: $0.06667/min | per minute | None stated | UNSPECIFIED | Baseten docs: instance types8 Nov 2026 |
| Runpod | NVIDIA GeForce RTX 509032 GB | ServerlessServerless worker (5090) | $1.79listed: $1.79/hr | per second | None stated | UNSPECIFIED | Runpod pricing8 Nov 2026 |
| Runpod | NVIDIA GeForce RTX 409024 GB | ServerlessServerless worker (4090) | $1.34listed: $1.34/hr | per second | None stated | UNSPECIFIED | Runpod pricing8 Nov 2026 |
Normalised to US dollars per GPU per hour from each provider's published price; "listed" shows the wording as published. "Contact sales" and "Not published" mean no public price was found. We never estimate one. Prices exclude storage, egress and taxes unless the row says otherwise.
“Serverless GPU” covers products that look alike but bill differently. All three platforms here start GPU workers on demand and scale to zero when idle, so you pay for running time, not for an always-on machine. What counts as running time, and what the rate includes, is where they differ.
Billing models side by side
| Modal | Runpod Serverless | Baseten | |
|---|---|---|---|
| Unit | Per second | Per second, rounded up | Per minute of instance time |
| GPU rate includes CPU/RAM? | No: CPU $0.0000131/core/sec and memory $0.00000222/GiB/sec on top | The tier rate covers the worker | Yes: each instance type bundles vCPU and RAM |
| Idle time billed | Containers stay up 60 s after the last input by default (configurable), billed | Idle wait up to the idle timeout (default 5 s), billed | Not stated on the instance-type page |
| Cold start billed | Billed from application load | Billed while loading the handler and models into GPU memory; image pull and cached-model download are not | Not stated on the instance-type page |
| Rate modifiers | Region selection 1.15–1.75x; non-preemptible 3x | Active (always-on) workers discounted via sales | None listed on the instance-type page |
| Included usage | $30/month compute on Starter | None stated | None stated on the instance-type page |
Implication. Modal’s listed GPU rates are the lowest of the three for most GPU classes in the index (Runpod’s L4 and L40S tiers are lower), but they are base rates for preemptible execution and exclude CPU and memory. Runpod’s and Baseten’s figures are closer to all-in per worker. Compare an estimate of the full bill, not the headline GPU rate.
When serverless beats a rented GPU
Serverless pays off when the GPU would otherwise sit idle: spiky inference, nightly batch jobs, internal tools used in working hours. When a worker would be busy most of the day, a reserved or on-demand GPU usually costs less per hour. The GPU Cloud Index shows both, and where to rent a GPU for AI walks through the trade-off.
What is not compared here
Performance figures such as cold-start seconds or throughput are not published consistently, and we do not run our own benchmarks. This comparison sticks to documented billing and product behaviour.
Provider facts
Modal
Serverless platform: Python functions and containers scale to zero and bill per second for GPU, CPU and memory.
- Billing
- per second
- Self-serve
- Yes
- Regions
- UNSPECIFIED
- Rate basis
- Listed GPU rates are base prices; non-preemptible execution is 3× base and region selection 1.15–1.75×
- Extras
- CPU and memory billed separately; idle containers billed until the scale-down window (60 s default)
- Free usage
- $30/month included compute on the Starter plan
Verified 8 Nov 2026 · Modal pricing
Runpod
GPU pods on Secure Cloud (Runpod-operated data centres) or Community Cloud (vetted third-party hosts), plus Serverless endpoints and multi-node Clusters.
- Billing
- per second
- Self-serve
- Yes
- Regions
- UNSPECIFIED
- Commitment
- None for Pods ("billed per second with no commitment"); reserved capacity via sales
- Storage
- Volume disk $0.10/GB/mo running, $0.20/GB/mo idle; network storage from $0.05/GB/mo
- Footprint
- "30+ regions" (no region list on the pricing page)
Verified 8 Nov 2026 · Runpod pricing
Baseten
Model-serving platform: dedicated deployments on chosen GPU instance types that autoscale, priced per minute of instance time.
- Billing
- per minute
- Self-serve
- Yes
- Regions
- UNSPECIFIED
- Rate basis
- Per-minute price per instance type (GPU + fixed vCPU/RAM bundle)
Verified 8 Nov 2026 · Baseten docs: instance types
Sources
- Modal pricing, checked 8 Nov 2026
- Runpod pricing (Serverless), checked 8 Nov 2026
- Runpod docs: Serverless pricing, checked 8 Nov 2026
- Baseten docs: instance types, checked 8 Nov 2026
Index rows carry their own source link and verification date. This page is revisited when: a platform changes per-second/per-minute GPU rates, CPU or memory charges, idle or scale-down billing, rate multipliers, or adds or removes a GPU class