AW3 - Multi-Cloud Hosting Platform
    Migrate · RunPod → Modal

    Migrate from RunPod to Modal
    handlers to functions.

    RunPod Serverless runs your Docker image with a Python handler; Modal runs Python functions with images defined in code. You can bring the same image and wrap the handler.

    Facts checked against each provider's own pages

    The short answer

    Bring your worker image to Modal with modal.Image.from_registry (or from_dockerfile)[1], wrap the handler in an @app.function exposed with @modal.fastapi_endpoint[2], move credentials into modal.Secret[3] and weights into a modal.Volume[4], deploy with modal deploy[5], then shift traffic.

    Mapping

    What maps to what

    On RunPodOn ModalNotes
    Serverless worker imagemodal.Image.from_registry / from_dockerfileWorks with Docker Hub, nvcr.io, ECR, ghcr.io and other registries[1].
    handler(job) + runpod.serverless.start@app.function + @modal.fastapi_endpointRunPod reads job['input'][6]; on Modal, fastapi_endpoint exposes the function over HTTP[2].
    Endpoint environment variablesmodal.Secret.from_nameInjected as environment variables via secrets=[...][3].
    Network volumemodal.Volume.from_nameMount with volumes={...}; call vol.commit() to persist writes[4].
    Flex workers, idle timeout 5 sScale to zero, scaledown_window 60 s[7][8]
    Active workers (always on)min_containersKeeps warm containers so the function never scales to zero[8].
    GPU tier per endpointgpu= on @app.functionModal lists B300, B200, H200, H100, A100, L40S, A10, L4 and T4 per second[9].
    Steps

    The migration, step by step

    1. 01

      Reuse the image

      Point modal.Image.from_registry at the image your RunPod endpoint already uses, or build from its Dockerfile with modal.Image.from_dockerfile[1].

    2. 02

      Wrap the handler

      Your RunPod handler takes a job and reads job['input'][6]. Call the same inference code from a Modal function decorated with @app.function and @modal.fastapi_endpoint, which gives it an HTTP URL[2].

    3. 03

      Move secrets and weights

      Create a Modal Secret for API keys and tokens[3]. RunPod network volumes don't migrate automatically, so copy model weights into a Modal Volume and commit them[4].

    4. 04

      Tune idle behaviour

      Modal keeps containers alive for 60 seconds after the last input by default and bills that time[9]; RunPod flex workers default to a 5-second idle timeout[7]. Adjust scaledown_window, or set min_containers to match RunPod active workers[8].

    5. 05

      Deploy

      Run modal deploy to publish the app and get the endpoint URL[5]. Test it with the same payloads your RunPod endpoint receives.

    Gotchas

    Things the docs warn about

    • Modal bills during application load time as well as input processing[9]; with large weights, cold starts are a cost line, not just latency.
    • Modal's GPU list prices are preemptible base prices. Non-preemptible runs cost 3×, and pinning a region multiplies prices by 1.15–1.75×[9].
    • Modal charges $0.04/GiB of egress beyond the plan allowance[9]; RunPod Pods have no egress fees[10] (its Serverless docs list no egress line item).
    • Request/response shapes differ: RunPod wraps payloads in a job with an input field; adapt clients, or accept the same shape in your Modal endpoint.
    Where AW3 Host fits

    Run both during cutover, then switch

    AW3 Host connects both your RunPod and Modal accounts, so the old and new deployments run side by side from one dashboard and API instead of a big-bang switch.

    Weighted routing sends a small share of traffic to Modal first (for example 5%, then 25%, 50% and 100%) while RunPod serves the rest. Health checks every 60 seconds watch both; if the Modal side fails, failover moves traffic back automatically, and rollback to the previous release is one step. Daily cost sync shows what each provider costs during the overlap, and budget policies alert you if the double-running period gets expensive. Secrets for both are stored encrypted per environment. Because both scale to zero, running them side by side during the overlap mostly costs the warm capacity you choose to keep.

    When Modal has carried all traffic cleanly for long enough that you no longer need the fallback, point DNS at it for good and decommission RunPod.

    Reference

    RunPod vs Modal at a glance

    RunPod vs Modal at a glance
    RunPodModal
    CategoryGPU cloudServerless GPU
    Pricing modelPay as you go: Pods, Serverless workers and Clusters billed by usage; storage per GB-month[11]Plan fee plus per-second usage for GPU, CPU cores and memory[9]
    Compute billing granularityPer second for Pods and Serverless (Serverless rounded up to the nearest second)[7]Per second (e.g. H100 $0.001097/s; CPU $0.0000131/core/s, minimum 0.125 cores)[9]
    Internet egress (first paid tier)No fees for data ingress or egress on Pods[10]$0.04/GiB beyond the plan allowance[9]
    Regions / locations31 global regions (also described as 30+)[11]9 broad and 11 narrow region codes for region selection (counted from list); pinning costs 1.15–1.75×[14]
    Deploy modelDocker images: Pods from templates; Serverless workers run your image with a Python handler[12]Python decorators define functions, images and endpoints; deploy with modal deploy[5]
    Persistent volumesYes: volume disks ($0.10/GB/mo running) and network volumes ($0.07/GB/mo under 1 TB)[13]Yes: Modal Volumes, $0.09/GiB/mo (1 TiB/mo free)[9]
    Config file—None: configuration lives in Python decorators[14]
    FAQ

    Frequently asked questions

    Can I use my RunPod Docker image on Modal?
    Yes. modal.Image.from_registry pulls from public or private registries such as Docker Hub, nvcr.io, ECR and ghcr.io, and modal.Image.from_dockerfile builds from a Dockerfile.
    What replaces RunPod active workers on Modal?
    min_containers, which keeps a floor of warm containers so the function never scales to zero. RunPod active workers, by comparison, run 24/7.
    How is idle time billed on RunPod and Modal?
    RunPod flex workers wait up to the idle timeout (5 seconds by default) before scaling down. Modal keeps containers alive 60 seconds after the last input by default and bills that time; both are configurable.
    Sources

    Sources

    1. Modal Docs: Existing images. https://modal.com/docs/guide/existing-images (checked 2026-11-16)
    2. Modal Docs: Web endpoints. https://modal.com/docs/guide/webhooks (checked 2026-11-16)
    3. Modal Docs: Secrets. https://modal.com/docs/guide/secrets (checked 2026-11-16)
    4. Modal Docs: Volumes. https://modal.com/docs/guide/volumes (checked 2026-11-16)
    5. Modal Docs: Managing deployments. https://modal.com/docs/guide/managing-deployments (checked 2026-11-16)
    6. RunPod Docs: Handler functions. https://docs.runpod.io/serverless/workers/handler-functions (checked 2026-11-16)
    7. RunPod Docs: Serverless pricing. https://docs.runpod.io/serverless/pricing (checked 2026-11-16)
    8. Modal Docs: Scaling out. https://modal.com/docs/guide/scale (checked 2026-11-16)
    9. Modal: Pricing. https://modal.com/pricing (checked 2026-11-16)
    10. RunPod Docs: Pod pricing. https://docs.runpod.io/pods/pricing (checked 2026-11-16)
    11. RunPod: Pricing. https://www.runpod.io/pricing (checked 2026-11-16)
    12. RunPod Docs: Serverless overview. https://docs.runpod.io/serverless/overview (checked 2026-11-16)
    13. RunPod Docs: Network volumes. https://docs.runpod.io/storage/network-volumes (checked 2026-11-16)
    14. Modal Docs: Region selection. https://modal.com/docs/guide/region-selection (checked 2026-11-16)
    Public beta

    Your next deploy
    goes everywhere.

    Connect a provider, point a service at it, and ship. Add the second provider when you're ready.

    No card needed · Free tier · Pay as you go after that