Modal
New York City, USA
AI infrastructure that developers love — the production cloud for AI.
What it is
Modal is a serverless cloud-compute platform for AI/ML — inference, fine-tuning, batch processing, and sandboxed code run from Python without managing servers. It bills per second and autoscales from zero.
At a glance
7/7 facts sourced- Pricing
- Starter is $0 base + $30/mo of free compute credits; Team is $250/mo base + $100/mo credits; Enterprise is custom. Usage is per-second (e.g. H100 GPU at $0.001097/s, CPU at $0.0000131/core/s).source ↗
- Hosting model
- Fully managed, multi-cloud serverless with a proprietary container runtime and scheduler that autoscales from zero to 1000+ GPUs.source ↗
- Memory & state
- Durable primitives — Volumes (a distributed file system persisted across invocations), plus Dicts and Queues. This is infrastructure state, not agent memory.source ↗
- MCP support
- Partialsource ↗
- Integrations
- A Python-native SDK and CLI as the primary interface, plus Sandboxes, Notebooks, web-endpoint deployment, and observability.source ↗
- Model providers
- Not applicable — pure compute; you bring and run your own models.source ↗
- Open source
- Nosource ↗
Editorial scorecard
Overall 3.5 / 5
Our opinion — the same 0–5 rubric is applied to every platform, including Alfe. See the rubric.
Strengths
- Per-second billing with recurring free credits and no idle charges.
- Python-native, autoscales from zero to many GPUs, with a broad GPU catalogue.
- Durable storage built in via Volumes, Dicts, and Queues.
Trade-offs
- The managed runtime is proprietary and not self-hostable.
- Higher concurrency and more seats require the Team ($250/mo) or Enterprise tier.
- Not an LLM provider or agent framework — you deploy and manage models yourself, and MCP means hosting your own server.
Our take
Editorial opinion — not a hands-on test resultIn our opinion, Modal is an excellent fit when you want serverless GPU/CPU compute for AI workloads with a first-class Python experience, and you are comfortable assembling the agent layer yourself. It shines on developer experience and scale; it is not, on its own, an agent platform, and MCP is a host-your-own affair rather than a built-in feature.
Sponsored
Alfe is a managed agent-hosting platform. If you want an agent hosted and running without wiring infrastructure yourself, see Alfe → (or read its dossier).
FAQ
- Is there a free tier?
- The Starter plan is $0 base plus $30/mo of free compute credits; the Team plan is $250/mo plus $100/mo in credits.
- Can Modal host an MCP server?
- Yes — you can deploy your own MCP server on Modal, but MCP is not a built-in platform feature.
Compare Modal
Sources
- Modal homepage · accessed 2026-07-01
- Modal pricing · accessed 2026-07-01
- Modal company page · accessed 2026-07-01
- Modal Volumes docs · accessed 2026-07-01
The Agent Examiner editorial · last updated 2026-07-01. Facts are drawn from each vendor’s own public materials; scores are our editorial opinion (methodology).
Product and company names are trademarks of their respective owners and are used here only to identify the products under discussion.