Modal
Choose Modal when you need to run your own AI code with explicit resource and execution control.
- Starting price
- Starter USD 0 fixed fee plus metered compute
- Free access
- USD 30/month compute credit on Starter
- Billing unit
- GPU seconds, physical-core seconds, GiB seconds and storage
- Platform
- Python SDK and managed cloud execution
- Verified
- 2 Nov 2026
Overview
What you provide
- Provide Python code, a container environment, model files and explicit CPU, memory or GPU requirements.
What you get
- Deployed functions, batch jobs or sandboxes return application results together with execution logs and resource usage.
Choosing this tool
It is infrastructure rather than a model-quality guarantee. A GPU line item is only one part of the bill: allocated CPU, memory and data movement can remain material. Replicate is simpler when calling an already packaged model; Together AI is relevant for a managed inference service with model-specific APIs. Modal is a useful fit for custom Python inference, batch jobs and agent sandboxes, but sandbox CPU and memory must not inherit the cheaper Functions estimate.
A decision worked through
At the listed base T4 rate, one GPU-hour is 3,600 × USD 0.000164 = USD 0.5904 before CPU, memory, storage, egress or regional premiums. This is an illustrative component cost, not a complete inference quote.
Modal alternatives for serverless compute for ai workloads and agent sandboxes
Compare Modal with Replicate, Together AI for this workflow, including the billing unit and the work left to your team.
| Product | Price / unit | Free allowance | Input → output | Key limit |
|---|---|---|---|---|
| ModalRead profileOfficial source ↗ | Starter USD 0 fixed fee plus metered compute | USD 30 compute credit each month | Python code and model assets become cloud jobs or endpoints. | Sandbox CPU/memory and regional execution use different rates. |
| ReplicateRead profileOfficial source ↗ | Per model: hardware seconds or input/output; FLUX 1.1 Pro USD 0.04/image | Some models can be tried; no universal perpetual allowance verified | Model API inputs return generated media or predictions. | Private deployments and public-model billing have different resource rules. |
| Together AIRead profileOfficial source ↗ | Model-specific token rates; gpt-oss-120B USD 0.15 input / USD 0.60 output per million tokens | Free rates exist for selected listed models; not a universal credit grant | Managed model requests return inference results through APIs. | Serverless, dedicated inference and GPU clusters are different products. |
Setup & workflow
- Deploy a minimal function with explicitly bounded resource requests.
- Run a representative input and inspect execution logs and metered usage.
- Set concurrency and spending controls before increasing workload volume.
Requirements & installation
- Prepare a Python environment and authorised cloud account; estimate Functions and Sandbox resources separately.
Limits & review
- Starter has one-day log retention and no deployment rollbacks in the comparison table. Included credits do not remove compute charges beyond the allowance.
Your part
- Validate model licensing, execution permissions and a representative resource bill before production scaling.
Sources & verification5
This profile is based on official sources, not a hands-on product test.
Vendor descriptions and demos document advertised features. Editorial guidance is based on these sources.
- Official Modal pricingmodal.com
- Official Modal productmodal.com
- Modal official referencemodal.com
- Replicate official referencereplicate.com
- Together AI official referencewww.together.ai
