Cloud Run vs Cloud Run Functions (2nd Gen) & Cloud Functions (1st Gen)

Date: 23 July 2025

1. Executive Summary

Google Cloud now offers three closely related serverless compute flavors:

Product	What it is	Typical Fit
Cloud Run service	Run any OCI container as an HTTP service or background job.	Full micro‑services, ML inference, batch jobs.
*Cloud Run function* (2nd Gen)**	Source‑only functions that compile down to a Cloud Run container under the hood.	Event‑driven glue code, quick HTTP endpoints.
Cloud Functions (1st Gen)	Legacy platform still billed on invocations/GB‑s/GHz‑s.	Existing workloads awaiting upgrade.

For sporadic traffic, Cloud Run request‑based billing and Cloud Run functions cost the same; migration changes developer experience more than the bill.

2. Terminology & Current Names

“Cloud Run functions” = Cloud Functions 2nd Gen (2024‑present).
“Request‑based billing” = pay ‑per‑vCPU‑second, ‑GiB‑second only while handling a request, plus $0.40 / million requests after free tier.
“Instance‑based billing” = traditional wall‑clock billing; CPU stays allocated between requests.

3. Feature & Configuration Differences

Dimension	Cloud Run service	Cloud Run function (2nd Gen)
Packaging	Bring your own container (Docker, Buildpacks, etc.).	Push source; Google builds the container.
Triggers	HTTP/gRPC by default; other events via Eventarc.	HTTP plus built‑in triggers (Storage, Pub/Sub, Firestore, Audit Logs…).
Concurrency	1 – 1 000 (80 default).	1 default; up to 1 000 configurable.
Max resources	8 vCPU / 32 GiB RAM (GPUs, volumes).	4 vCPU / 16 GiB RAM.
Traffic splitting	Full rollout/rollback, A/B.	Available (not in 1st‑Gen).
Billing modes	Request‑based or instance‑based.	Request‑based only.

4. Pricing Models (USD list, `us‑central1` baseline)

4.1 Cloud Run request‑based (applies to services & functions)

Component	Rate	Free / mo
Active CPU	$0.000024 per vCPU‑s	180 000 vCPU‑s
Idle CPU (min instances)	$0.0000025 per vCPU‑s	—
Active Memory	$0.0000025 per GiB‑s	360 000 GiB‑s
Idle Memory (min inst.)	$0.0000025 per GiB‑s	—
Requests	$0.40 per M	2 M

4.2 Cloud Run instance‑based (services only)

Component	Rate	Free / mo
CPU	$0.000018 per vCPU‑s	240 000 vCPU‑s
Memory	$0.000002 per GiB‑s	450 000 GiB‑s
Requests	None	N/A

4.3 Cloud Functions 1st Gen (legacy)

Component	Rate	Free / mo
Invocations	$0.40 per M	2 M
Memory	$0.0000025 per GB‑s	400 k GB‑s
CPU	$0.0000100 per GHz‑s	200 k GHz‑s
Internet egress	$0.12 / GB	5 GB

5. Cost Illustration – Sporadic Workload

Scenario

10 k invocations / month
500 ms execution each
1 vCPU, 256 MiB
Scales to 0 between bursts
Region: europe‑west2

Model	Calculated Monthly Cost
Cloud Run function (2nd Gen)	$0 (all within free tier)
Cloud Run service (request‑based)	$0
Cloud Run service (instance‑based)	~$0.09 if instance stays up the whole 5 s total; higher if kept warm

Result: No cost advantage to migrating when the service already uses request‑based billing.

6. When Migration Does Make Sense

You need built‑in Cloud event triggers without managing Eventarc.
You like source‑only deployments (skip Docker).
You want strict 1‑request isolation by default.
You are retiring 1st‑Gen Cloud Functions and want lowest‑effort port.

7. Migration Checklist (Service → Function)

Map the entry point – Export a single handler (main.py, index.js, etc.).
Remove unnecessary endpoints/background loops.
Adjust config – Env vars, memory ≤ 16 GiB, vCPU ≤ 4, timeout ≤ 60 min.

Deploy

gcloud run deploy my‑function        --source=.        --region=europe-west2        --trigger-http        --allow-unauthenticated

Add event triggers if required.
Update clients/DNS – new URL.

Rollback is instant via previous Cloud Run revision.

8. References

Google Cloud Run & Cloud Run Functions pricing docs, retrieved 23 July 2025.
Google Cloud product release notes, Aug 2024 – Jul 2025 (Cloud Functions 2nd Gen GA).
Google Cloud blog “Cloud Run pricing updates: request‑based vs instance‑based modes”, 14 Mar 2024.

Cloud Run vs Cloud Run Functions (2nd Gen) & Cloud Functions (1st Gen)