by Modal Labs

Modal pricing, free plan and limits

AI infrastructure that developers love: serverless compute for ML inference, training, and batch processing

Best for
ML Engineer (need GPU access without Kubernetes complexity)
Not for
Enterprise DevOps (limited customization, no BYOC)

Should you use Modal?

Modal suits ML engineers and startups who want GPU compute without touching Kubernetes: write a normal Python function, add one decorator, and Modal handles the container, scaling, and orchestration, with cold starts landing in under 1 second. It replaces hand-rolled Docker and Terraform pipelines for teams that would rather ship models than maintain infrastructure themselves.

Sub-second cold starts combined with Python-first infrastructure—developers can write serverless functions like normal code and scale to hundreds of GPUs without DevOps overhead

Modal has built proprietary infrastructure (custom container runtime using gVisor, VolumeFS, memory snapshots, multi-cloud orchestration) that is difficult to replicate. The developer experience moat (Python decorators, local-like iteration) is strong but replicable. Network effects are emerging as enterprises standardize on the platform. However, the Python-only positioning and lack of BYOC limit total addressable market compared to generalist platforms like AWS Lambda.

Also for

  • Data Scientist (need to scale local workflows to cloud instantly)
  • Research Team (rapid experimentation with models)
Show 8 more
  • Startup CTO (minimal DevOps overhead, fast deployment)
  • ML Platform Engineer (building internal AI infrastructure)
  • ML Engineer
  • Data Scientist
  • Backend Developer
  • DevOps Engineer
  • Startup CTO
  • Research Team

Also not for

  • Polyglot Backend Engineer (Python-first limits flexibility)
  • On-premise-first organization (managed-only deployment)
  • Team requiring strict data residency compliance (limited to specific regions)

Main trade-off

Python-only ecosystem and vendor lock-in with managed-only deployment model; limited for polyglot teams or those requiring on-premise/BYOC options

Modal pricing

PlanMonthly priceWhat it includes
StarterFreePer-second CPU/GPU/Memory usage with $30/month free credits; 100 containers max, 10 GPU concurrency
EnterpriseFreeCustom pricing with volume-based discounts, usage-based compute on top of platform fee, unlimited seats, higher GPU concurrency, dedicated support
Team$250/moPer-second CPU/GPU/Memory usage with $100/month in credits on top of $250/month subscription; 1000 containers, 50 GPU concurrency, unlimited seats
How the pricing works, in full

Free Starter tier with $30/month in compute credits. Team plan at $250/month (includes $100/month in credits, unlimited seats, 1000 containers, 50 GPU concurrency). Enterprise plan with custom pricing for volume-based discounts and higher concurrency. On top of subscription fees, usage is pay-as-you-go: CPU at $0.0000131/core/sec (min 0.125 cores), Memory at $0.00000222/GiB/sec, GPUs ranging from $0.000164/sec (T4) to $0.001097/sec (H100). Multipliers apply for regional selection and non-preemptible sandboxes.

Before you commit

  • Developer hours for optimization (cold starts still require thoughtful architecture despite improvements)
  • Regional multipliers (1.25x-2.5x) and non-preemptible sandbox multipliers (3x) can significantly increase compute costs beyond base rates
  • Model loading latency (1-5 minutes for multi-gigabyte models on first cold start, even with container optimization)

What Modal does

Modal runs containers up to 100x faster than Docker, giving AI teams serverless compute for inference, training, and batch jobs without managing servers. It automatically scales from zero to hundreds of GPUs across multiple clouds, targeting ML teams that need infrastructure to expand and shrink with demand rather than staying provisioned around the clock.

  • Sub-second Cold Starts

    Spin up GPU-enabled containers in as little as one second with custom infrastructure optimized for rapid iteration and scaling, eliminating cold start latency that plagues traditional serverless platforms.

  • Elastic GPU Scaling

    Autoscale from zero to hundreds of GPUs on-demand based on workload, with deep multi-cloud capacity and intelligent scheduling that ensures access to CPUs and GPUs without managing quotas or reservations.

  • Python-First Infrastructure

    Define entire environments, hardware requirements, and deployment logic in pure Python using simple decorators (@app.function, @app.cls) without YAML files or complex configuration, keeping environment and hardware requirements in sync.

  • Unified Observability

    Integrated logging and full visibility into every function, container, and workload with built-in dashboards, real-time logs, and first-party integrations with Datadog and OpenTelemetry for monitoring and debugging.

Show all 6 features
  • Multi-workload Support

    Deploy inference for LLMs and generative models, fine-tune open-source models, run large-scale batch jobs, execute untrusted code in sandboxes, and collaborate in real-time with shareable notebooks, all from a unified platform.

  • Memory Snapshots

    Dramatically reduce cold start latency (up to 10x) by capturing container state after initialization and restoring it for subsequent starts, enabling GPU snapshots for even faster LLM deployments.

Works with

  • AWS S3 (cloud bucket mounts)
  • Google Cloud Storage (cloud bucket mounts)
  • Datadog (observability and monitoring)
  • OpenTelemetry (logging and tracing)
  • FastAPI (web endpoints)
  • Hugging Face (model hub integration)
  • Stripe (payment processing examples)
  • GitHub Actions (CI/CD deployment)
  • Slack (notifications and logging)

Missing integrations

  • Kubernetes (Modal is alternative, not integration)
  • Terraform/CloudFormation (infrastructure-as-code for Modal itself)
  • Bring Your Own Cloud (BYOC) deployment option
  • Non-Python SDK (Go, TypeScript have limited support)
  • VPC/Private network support (limited networking options)

Most users expect these and they are not currently supported.

Capabilities

  • Inference
  • Training
  • Batch processing
  • Serverless
  • Container orchestration
  • GPU acceleration
  • LLM deployment
  • Model fine tuning

Strengths and limits

Strengths

  • Exceptional developer experience with Python decorators, minimal boilerplate, and local-like iteration loops in the cloud
  • Cold starts are quick enough that GPU functions behave like ordinary local code instead of waiting through a container boot, which matters most for interactive inference
  • Flexible usage-based pricing without per-request charges, scaling from zero to thousands of GPUs on-demand
  • Genuinely serverless: no infrastructure management, no Kubernetes, no container orchestration to maintain
  • Deep multi-cloud capacity with intelligent scheduling ensuring GPU access without managing quotas or reservations

Limits

  • Python-first platform limits flexibility for polyglot teams or non-Python workloads
  • Vendor lock-in due to managed-only deployment model (no Bring Your Own Cloud option)
  • Less granular control over underlying infrastructure compared to traditional cloud providers like AWS/GCP
  • Regional limitations for latency-sensitive inference (primarily Ashburn, Virginia for HTTP traffic)

Modal alternatives

ToolPrice fromFree planRuns onRating
Modal This page$250/moYesWeb--
ReplicateReplicate covers calling a pre-packaged open-source model over an API and nothing more. Modal is built for the opposite case: running and scaling custom Python code you wrote yourself.FreeYesWeb--
Together AITogether AI works from a managed catalog of ready-to-fine-tune open-source models. Modal has no catalog at all, built instead for custom Python code and container logic that goes beyond any fixed model list.--TrialWeb--
Fireworks AIFireworks AI is scoped to pay-per-token LLM inference alone. Modal covers more ground: training, batch jobs, and sandboxes, all billed per second instead of per token.--YesWeb4.4

Also compared

  • GroqGroq's LPU hardware delivers low-latency inference, but only across a fixed model lineup. Modal runs general-purpose GPU compute instead, covering training, fine-tuning, and custom inference code Groq's fixed lineup can't touch.

Modal facts

Works offline
No
Minimum skill level
Intermediate
Price band
Mid-range
Maker founded
2021
Total funding
$111M Series B (July 2025)
Compliance
SOC 2 · HIPAA · Okta SSO · audit logs · data residency controls

Architecture and specs

PlanArchitecture
StarterServerless container orchestration
TeamServerless container orchestration with advanced features
EnterpriseServerless container orchestration with enterprise controls

Data and privacy

MIT (modal-client), Apache-2.0 (libmodal)

What changed

  • New feature Modal serverless compute adds GPU autoscaling for inference workloads source

Questions about Modal

What does Modal actually cost?

Modal has three tiers: Starter is free, Team costs $250 a month, and Enterprise is custom-priced with volume discounts. Beyond the included credits, usage is billed per second for CPU, memory, and GPU time, with extra multipliers applied for regional selection or non-preemptible sandboxes.

What do you get on Modal's free tier?

Modal's free Starter tier includes $30 a month in compute credits, capped at 10 concurrent GPU workers. It does not include the Team plan's unlimited seats or 50-GPU concurrency, and any usage beyond the included credits bills at the same per-second rates as the paid tiers.

Which tools compete with Modal in 2026?

Replicate is the simpler choice for calling a pre-packaged open-source model through an API rather than writing custom deployment code. Together AI offers a managed catalog of open-source models to fine-tune without building a container pipeline from scratch. Fireworks AI is worth a look when the need is pay-per-token LLM inference rather than general-purpose Python compute.

Modal or AWS Lambda: which should you pick?

Modal is purpose-built for ML workloads, offering direct GPU access and support for long-running jobs that Lambda cannot handle. AWS Lambda caps out at a 15-minute timeout and a 50MB deployment package, making it a fit for short, event-driven glue code rather than model inference or training. Choose Modal for AI workloads and Lambda for lightweight application backends.

How long does it take to get going with Modal?

A first deployment usually takes minutes: install the Python client and wrap any existing Python function with the @app.function decorator instead of writing a Dockerfile or YAML config, and Modal handles containerization and scaling automatically. New accounts default to the free Starter tier, so early deployments cost nothing until the included credits run out. The official Getting Started guide at modal.com/docs/guide walks through that first deployment step by step.

About Modal

Modal Labs, founded in 2021, raised a $111M Series B in July 2025. The company built a serverless compute platform purpose-built for AI, ML, and data teams: developers run compute-intensive workloads without managing infrastructure, with sub-second container cold starts, instant autoscaling, and a developer experience that feels local. Define everything in code using Python decorators (no YAML, no Dockerfiles) and Modal handles containerization, resource allocation, scaling, and orchestration automatically.

Read the full background

The platform is engineered from the ground up for heavy AI workloads, with multi-cloud GPU capacity, globally distributed storage, and integrated observability. Whether deploying LLM inference at scale, fine-tuning models on multi-node clusters, running massive batch jobs, or executing untrusted code in sandboxes, Modal abstracts away operational complexity so teams can focus on model development rather than infrastructure management.

Sources and guides

More AI Tools on Hokai

  • KinoviKinovi is a pay-as-you-go API and browser studio that runs dozens of video, image and audio models from ByteDance, Alibaba, xAI and others on one credit balance.
  • Tamarind BioHosted platform for AlphaFold, RFdiffusion, Boltz and hundreds of other biology tools via web app, API and MCP, with a Free plan to start on.
  • ArchilA durable shared disk for AI agents that sits on top of your object storage, with serverless sandboxes that bill per tool call and a free tier.
  • Google AI StudioFree web-based IDE for building, testing, and deploying generative AI applications with Google's Gemini models
  • Cast.aiCast.ai cuts Kubernetes cloud costs 50-70% using automated rightsizing and Spot instance management. Trusted by 2,100+ companies, reached $1B unicorn in 2026.
  • SupermicroSupermicro designs AI server hardware from edge systems (65W TDP) to 8-GPU data center nodes, with $12.68B quarterly revenue and NASDAQ: SMCI listing.