What GCP Services Are Free Tier in 2026?

What GCP Services Are Free Tier in 2026?

You’re building something. Maybe it’s a prototype for a startup, a side project that could blow up, or a data pipeline you want to test without asking for a budget. The question everyone asks: “What GCP services are free tier?” I’ll answer that — but I’ll also tell you where the traps are.

I’ve been running production systems on GCP since 2019. At SIVARO, we’ve migrated clients from AWS to GCP, from Azure to GCP, and I’ve watched teams burn through credits because they assumed “free tier” meant “unlimited.” It doesn’t.

Here’s the honest breakdown: what’s actually free, what’s free for a limited window, and what looks free but will cost you if you blink.

The Three Tiers of Free (Most People Miss This)

Google Cloud’s free offerings come in three flavors. Ignore this and you’ll get a surprise bill.

1. Always Free — never expires, but limited in capacity.
2. 12-Month Free Trial — $300 credit, usable on any service, expires after a year.
3. Limited Free Quotas — some services offer a set amount of usage per month, regardless of trial status.

Most guides lump them together. That’s dangerous. I’ve seen a team spin up a high-memory VM thinking it was covered by the trial, then blow $800 in two weeks because they didn’t realize the trial credit only covers standard machine types.

Let’s start with the Always Free tier — the stuff that actually costs $0 indefinitely.

Compute Engine – The Always-Free Workhorses

Google offers two f1-micro instances per month (0.25 vCPU, 0.6 GB memory) plus 30 GB of persistent disk HDD, plus 1 GB of snapshots. That’s enough to run a small web server, a bot, or a dev database.

Is it fast? No. f1-micro is throttled — you get a CPU burst for a few seconds, then it chokes. I’ve run Mastodon instances on it for testing. Works fine for low traffic. Don’t try to run a production API on it.

bash
# Create an f1-micro instance in the free tier region (us-west1, us-central1, us-east1)
gcloud compute instances create my-free-instance     --machine-type=f1-micro     --zone=us-west1-b     --image-family=debian-12     --image-project=debian-cloud     --boot-disk-size=10GB     --boot-disk-type=pd-standard

That’s 10 GB of persistent disk — within the 30 GB free limit. You can have two of these. But remember: the 30 GB disk is shared across all your instances. If you attach a second disk, that counts.

Also: network egress is not free. A few GB of outbound traffic per month won’t kill you, but start streaming video and you’ll see charges. The free tier includes 1 GB of egress per month between GCP regions (except China/Australia). For internet egress, there’s no free allowance — you pay after the first 1 GB.

Cloud Storage – More Than You Think

The free tier for Cloud Storage is generous: 5 GB of regional storage (US-Central, US-East, US-West) per month, plus 1 GB of network egress to North America, plus 5,000 Class A operations and 50,000 Class B operations.

That’s enough for a small file server or a static website. Use it as a CDN origin, host images, or store logs. The catch: if you set your bucket to multi-regional, the free tier doesn’t apply. Stick with regional (default) to stay at zero cost.

bash
# Create a bucket in the free tier
gcloud storage buckets create gs://my-free-bucket     --location=US-CENTRAL1     --default-storage-class=STANDARD     --public-access-prevention

Want to serve static content? Pair it with Cloud Load Balancing — that’s not free, but the first 5 TB of egress from Cloud CDN is $0.08/GB, and if you’re small, it’s negligible. Or just use Cloud Storage directly. Works fine for a personal site.

Cloud Functions – Serverless on a Shoestring

Google offers 2 million invocations per month, plus 400,000 GB-seconds of compute time, plus 200,000 GHz-seconds, plus 5 GB of internet egress for free under the Always Free tier.

This is massive for a lot of use cases. I run a webhook receiver that processes ~500,000 events per month — zero cost. The catch: cold starts on the 128 MB tier can be 2–3 seconds. If latency matters, you’ll need to keep at least one instance warm (costs money) or use Cloud Run with min-instance = 0 (still free tier).

python
# Example Cloud Function (2nd gen) that fits in free tier
import functions_framework
import json

@functions_framework.http
def hello_free_tier(request):
    """HTTP Cloud Function."""
    request_json = request.get_json(silent=True)
    name = request_json.get('name', 'world')
    return f"Hello {name}! This function costs $0 if under 2M calls/month."

But watch your memory. The free tier compute time is based on memory-usage × duration. If you allocate 256 MB, your 400,000 GB-seconds evaporate twice as fast. Stay at 128 MB for most functions.

Cloud Run – Containerized but Free (Sort Of)

Cloud Run has a similar Always Free tier: 2 million requests per month, 360,000 GB-seconds of compute time, 180,000 vCPU-seconds, and 1 GB of egress.

That’s enough to serve a small API. I’ve deployed a FastAPI app serving a tiny model — 50 requests per second for maybe 10 seconds a day — zero bill. The gotcha: you’re billed for vCPU-seconds even when idle if you set min-instances > 0. Keep min-instances = 0, and you only pay while a request is being processed.

BigQuery – Free Queries, Not Free Everything

BigQuery’s free tier is 1 TB of query data processed per month, plus 10 GB of storage. That’s plenty for exploratory analysis, small datasets, or dashboards.

But here’s where people get burned: streaming inserts. Each row inserted via streaming costs $0.01 per 200 MB (roughly). If you stream 1 GB, that’s $0.05 — tiny, but it adds up. The free tier doesn’t cover streaming inserts. Also, exporting data costs egress.

I’ve seen startups that built their whole analytics on BigQuery free tier, then hit 1 TB in a month because they ran a few full-table scans. The solution: use partitioning and clustering, and limit query size with a --maximum_bytes_billed flag.

sql
-- Set a query cost limit in the console or via API
-- This query will fail if it exceeds 1 GB processed
SELECT COUNT(*) FROM `bigquery-public-data.samples.gsod`
WHERE station_number = 123456

Set your default billing cap to 1 TB at the project level. You’ll get an email when you hit 50%, 90%, etc. Not a surprise.

Firestore and Firebase – Mobile Backend Without the Bill

Firestore’s free tier is 1 GB stored, 50,000 reads per day, 20,000 writes per day, 20,000 deletes per day. That’s enough for a small app with maybe 100 daily active users.

Real story: I used Firebase Auth + Firestore for a prototype I built in 2022. It handled 500 signups before I hit the write limit. Upgrading to Blaze (pay-as-you-go) cost me $0.18 that month. Not a big deal, but if you’re running a cron job that writes logs every minute, you’ll hit 20,000 writes in 14 days.

The Firebase Authentication free tier includes 10,000 MAU for phone auth, unlimited for email/password. That’s huge. Most identity providers charge per MAU.

But watch out for Cloud Functions triggered by Firestore changes. Those functions count toward your Cloud Functions free tier separately — but they also trigger more reads/writes. Chain them and you can eat through both allowances quickly.

Other Notable Free Services

• Cloud Pub/Sub: 10 GB of messages per month, plus 5 GB of egress. Perfect for event-driven architectures.
• Cloud Scheduler: 3 jobs per month. Not many, but enough for a cron backup.
• Cloud Monitoring: 10 metrics per month, plus 250 MB of logs ingestion. The alerting policy is limited but works.
• Cloud Source Repositories: 1 repo with 5 GB storage.
• Cloud Translation – Basic: 500,000 characters per month (for Basic tier only).
• Cloud Vision API: 1,000 units per month.

These are all Always Free. Link them together and you can build a decent pipeline. I’ve run a prototype for a retail analytics tool using Pub/Sub + Cloud Functions + Firestore — all free for the first 10,000 events per day.

What’s NOT Free? (And What’ll Burn Your Wallet)

The $300 trial credit is useful, but it’s a trap if you don’t know the exceptions.

• GPU instances: Not covered by Always Free. Even with trial credits, a GPU VM can burn $300 in a week.
• Dedicated Interconnect: Nope.
• Cloud SQL: The Always Free only applies to the f1-micro instance for Cloud SQL? Wait — no, Cloud SQL doesn’t have an Always Free tier. The trial credit covers it, but after that it’s $8–$200/month.
• App Engine: The standard environment has a free tier (28 instance-hours per day), but the flexible environment does not.
• Cloud Load Balancing: Charged per hour. Minimum $18/month for a global HTTPS load balancer.
• Network Egress: This is the silent killer. GCP charges $0.12/GB for most regions to internet. If your app serves files, videos, or APIs to users outside GCP, your bill will be dominated by egress. The free tier gives you 1 GB/month for egress — that’s nothing.

Compare: AWS’s free tier gives 1 GB of egress per month too. Azure gives 15 GB. Google is stingier here. See the Cloud Pricing Comparison 2026 — GCP is competitive on compute but egress is where they make margin.

GCP Free Tier vs AWS Free Tier vs Azure

AWS offers 750 hours of t2.micro (1 vCPU, 1 GB) for 12 months, plus 5 GB of S3 storage, 25 GB of DynamoDB, and 1 GB of egress. Azure offers 750 hours of B1s (1 vCPU, 1 GB) for 12 months, plus 5 GB of blob storage, 250 GB of SQL Database, and 15 GB of egress.

GCP’s Always Free (no expiry) f1-micro (0.25 vCPU, 0.6 GB) is weaker. But GCP’s 12-month trial gives $300 credit vs AWS’s 750 hours of compute (which is roughly $40–$60 of compute). GCP’s $300 is more flexible — you can spend it on GPUs, BigQuery, or networking.

For startups, the GCP vs AWS 2026 comparison shows GCP’s free tier is better for data-heavy workloads (BigQuery, Firestore) but worse for compute-heavy ones. AWS’s t2.micro can actually run a small production service; f1-micro struggles.

I’d say: if you’re building a serverless app with low traffic, GCP’s Always Free is unbeatable. If you need a real server for a production app, AWS free tier gives you more horsepower.

Is Google Cloud Platform Good for Startups?

Yes — but with caveats. The free tier gives you enough to validate a product. The $300 credit lets you test GPUs or BigQuery without upfront cost. And the Comparing AWS, Azure, and GCP for Startups in 2026 article notes GCP’s startup program (Google for Startups) offers up to $200,000 in credits over two years if you’re accepted.

But the pricing model is complex. GCP charges per resource (CPU, memory, disk) rather than per instance. If you forget to delete a persistent disk attached to a terminated VM, you’ll keep paying. AWS and Azure have similar traps, but GCP’s billing breakdown is less intuitive.

I’ve seen a startup rack up $2,000 in a month because they left a few high-memory preemptible instances running over a weekend. Preemptible VMs are cheap ($0.01–$0.02/hour), but if you launch 20 of them for 72 hours, that’s $30–$60. Not catastrophic, but they thought it was free because they were using spot pricing. Spot pricing is not free.

How to Migrate from Azure to GCP (and Save Costs)

If you’re reading this and thinking about switching, the migrate from azure to gcp guide I helped write at SIVARO covers the mechanics. For free tier specifically:

1. Map Azure’s free services to GCP equivalents. Azure’s App Service (F1) gives 1 GB memory and 1 GB disk — GCP’s Cloud Run free tier is more generous in requests but less in memory (128 MB per container). For a small app, Cloud Run is fine; for a larger one, you might need a paid tier.
2. Azure gives 250 GB of SQL Database free (12 months). GCP has no free managed SQL — you’ll need to run your own on Compute Engine or use Firestore.
3. Azure’s free egress (15 GB) is better than GCP’s (1 GB). If your app sends data to users, Azure wins.

But GCP’s BigQuery free tier (1 TB/month) is a killer feature for analytics. If you’re running a data-heavy workload, migrating from Azure to GCP can save you thousands.

Most people think moving clouds is about rehosting VMs. It’s not. It’s about rethinking the architecture to exploit each platform’s free tiers. At SIVARO, we migrated a logistics startup from Azure to GCP last year. They saved 40% on their monthly bill, partly because we moved their analytics from Postgres to BigQuery free tier. That alone covered the migration cost.

Tips to Stay in Free Tier (Without Surprise Bills)

1. Set budgets and alerts. In GCP console, create a budget for $1 and set alerts at 50%, 90%, 100%. Even if you think you’re safe, do this. I do it for every project, including personal ones.

2. Use the Pricing Calculator Google Cloud Pricing Calculator before launching anything. Simulate your expected usage. It’s not perfect — it underestimates egress — but it catches egregious errors.

3. Delete resources. A stopped Compute Engine VM still charges for attached disk. A Cloud SQL instance you’re not using still charges for storage. I’ve written a cleanup script that runs weekly via Cloud Scheduler.

bash
# Example cleanup script – list all VMs and disks not in use
gcloud compute instances list --format="value(name, zone)"
gcloud compute disks list --format="value(name, zone, users)"

4. Monitor egress. The free tier gives 1 GB. Use a custom dashboard in Cloud Monitoring to track egress costs. If you exceed 0.5 GB in a week, investigate.

5. Use preemptible VMs for batch jobs. They’re not free, but they’re 60–80% cheaper. Just make sure your job is fault-tolerant.

6. Avoid streaming inserts into BigQuery. Use batch loads. Those are free (within the 1 TB/month query limit). Streaming costs $0.01 per 200 MB — fine for occasional use, brutal for high volume.

FAQ – What GCP Services Are Free Tier?

Q: Can I run a production web app on GCP free tier?

Short answer: No. The f1-micro VM will choke under any real load. Cloud Run’s free tier (2M requests/month) can handle a small API if traffic spikes are low. For a production app, expect to pay at least $10–$30/month for a decent VM.

Q: Does the free tier include Cloud Load Balancer?

No. It’s always paid. The cheapest option is a regional external HTTPS load balancer at $0.025 per hour, plus $0.007 per GB of data processed. About $18/month minimum.

Q: What happens after the 12-month trial expires?

You lose the $300 credit. Any Always Free services remain free. Services you started during the trial will continue to be billed. The trial doesn’t auto-stop — you need to manually delete or downgrade resources.

Q: Is BigQuery free tier really 1 TB per month?

Yes — but that’s 1 TB of data processed by queries, not storage. You also get 10 GB of storage free. If you store 100 GB, you pay $0.02/GB/month after the first 10 GB.

Q: Can I combine the free tier with the $300 trial?

Yes. The Always Free services don’t consume your $300 credit. The credit is separate. You can use it for services not covered by Always Free (like GPUs, load balancers, or extra storage). But once the $300 runs out, you’ll be billed.

Q: Which GCP region is best for free tier?

The Always Free services are tied to specific regions: US-West1 (Oregon), US-Central1 (Iowa), US-East1 (South Carolina). For Cloud Storage, those three are the only ones offering free storage. For Compute Engine, you can choose any zone in those three regions.

Q: Is Firebase free tier the same as GCP free tier?

Partially. Firebase’s free tier (Spark plan) includes Firestore, Authentication, and Cloud Storage. It shares the same quotas as GCP’s Always Free. But Firebase Cloud Functions are separate — they use Cloud Functions quotas. So you can run both within the same project.

Q: How do I check my usage against free tier?

Go to the GCP Free Tier Dashboard. You’ll see a real-time meter for each service. Also enable billing exports to BigQuery for detailed analysis.

Final Word

The path to building something for free on GCP isn’t about getting lucky — it’s about knowing the boundaries. When I show startup founders what GCP services are free tier, most are surprised by the serverless compute limits. They assume “free” means “unlimited functions.” It doesn’t.

But the real value of GCP’s free tier isn’t saving $10 a month. It’s that you can prototype an entire product without asking for a credit card. I’ve built three prototypes this year alone using only Always Free: a webhook aggregator, a sentiment analysis pipeline, and a real-time dashboard. All cost $0.

Eventually you’ll need to pay. And that’s fine. The free tier gives you room to fail fast, learn, and then spend money exactly where it matters.

At SIVARO, we help companies escape the hidden costs of cloud — whether that’s migrating from Azure to GCP or just optimizing $50/month projects. But the fundamentals start with understanding the free tier. Because if you can’t build a prototype for free, you’re not ready to build it for real.

---

Nishaant Dixit — Founder of SIVARO. Building data infrastructure and production AI systems since 2018. Built systems processing 200K events/sec.