Google Cloud Storage pricing is GiB per month by storage class and location, plus operations per 1,000 by class, retrieval per GiB for the colder classes, early deletion for objects removed before the class's minimum duration, and egress per GiB by destination; Autoclass adds a fee per 1,000 objects, soft delete keeps deleted data billable for its retention window, and a dual-region bucket pays both regions. On the bill, Standard storage dominates for most buckets, but the colder classes hold the surprises: a Nearline bucket read daily pays more in retrieval and Class B operations than it saves on storage, and an Archive object deleted after a month pays the full year.
Pricing dimensions
A bucket's class and location set the rate per GiB; what the bucket is used for sets everything else. Every combination of class, location and operation kind has its own SKU, and the export names it by class and location.
| Dimension | Unit | What drives it | Optional |
|---|---|---|---|
| Storage | GiB per month by class (Standard, Nearline, Coldline, Archive) and location (region, dual-region, multi-region), by the second; dual-region billed to both regions | Stored bytes per class, soft-deleted and noncurrent versions included | No |
| Minimum storage duration | 30 days Nearline, 90 days Coldline, 365 days Archive; an object deleted, overwritten or moved earlier pays the remainder as early deletion | Objects that leave a cold class early, versioning and soft delete included | No in the cold classes |
| Class A operations | Per 1,000: writes, lists, inserts, rewrites; rates rise by class, Archive the dearest | Uploads, listings, lifecycle transitions | No |
| Class B operations | Per 1,000: reads and metadata gets; rates rise by class | Every object read, every metadata lookup | No |
| Retrieval | Per GiB read from Nearline, Coldline and Archive, on top of the operation | Reads from cold classes | No in the cold classes |
| Egress | Per GiB to the internet in tiers per month by destination; between Google Cloud locations by continent pair; within a region free; to other Google Cloud services in the same location free | Downloads, replication, cross-region reads | Yes |
| Autoclass | Management fee per 1,000 objects per 30 days; objects under 128 KiB unmanaged; no retrieval or early deletion fees on transitions | Buckets with Autoclass enabled | Yes |
| Soft delete | Deleted data billed at its class rate for the retention window, seven days by default, off or up to 90 days | Every deletion and overwrite in a bucket with soft delete | Yes |
| Turbo replication and Rapid storage | Per GiB for 15-minute dual-region replication; Rapid storage as a zonal class with its own rate per GiB and operation | Low-latency replication, zonal high-performance buckets | Yes |
| Storage Intelligence and insights | Per million objects per month for the object management fee, per GiB for datasets | Buckets enrolled in the subscription | Yes |
| Purchase option | None; no commitment for storage |
The class is a bet on access: Nearline costs about half of Standard per GiB, Coldline a quarter, Archive a tenth, and each step down raises the operation rates and adds a retrieval fee per GiB and a minimum duration. An object that is read once a month is cheaper in Nearline; one read daily is cheaper in Standard; one that will be deleted within a year is dearer in Archive than in Coldline, because the minimum duration is paid whether the object stays or not. Autoclass moves objects between classes by their access pattern and charges a fee per 1,000 objects instead of retrieval and early deletion; it pays off on buckets with many large objects and unknown access, not on buckets of small files, which are not managed below 128 KiB. Soft delete, on by default since 2024 with seven days, keeps every deleted or overwritten byte billable for the window, which on a bucket that rewrites its contents daily multiplies the storage line; the retention can be shortened or switched off per bucket. The mechanics across all three providers are in block, object and file storage and storage tiers and lifecycle.
How it appears in your billing data
In the billing export every row carries service.description Cloud Storage and a sku.description that names class and location: Standard Storage Frankfurt, Nearline Storage Frankfurt, Coldline Storage Frankfurt, Archive Storage Frankfurt for the bytes, Standard Class A Operations Frankfurt and Standard Class B Operations Frankfurt per 1,000, Nearline Data Retrieval Frankfurt and Coldline Data Retrieval Frankfurt per GiB, Early Delete Nearline Frankfurt for the minimum duration, Autoclass Management Fee, Networking Traffic Egress GCP Inter Region and Premium Tier Internet Data Transfer Out for the traffic; usage.unit is gibibyte-month for storage, count for operations and gibibyte for retrieval and egress. location.region is europe-west3 for a regional bucket and the pair or the continent for dual- and multi-region buckets, and resource.name in the export with resource-level detail names the bucket; without it, the bucket is visible only through labels, which a bucket carries and its objects do not. Soft-deleted and noncurrent bytes sit in the same storage SKU as live ones, so the export shows the total, and the bucket's own metrics separate live from soft-deleted bytes.
Credits are rare on this line: the free allowance of the always-free programme, 5 GB of Standard storage in US regions, appears as credits.type FREE_TIER, and there is no sustained use or committed use discount on storage. The export's structure is described in the Google Cloud billing export; in the FOCUS export, ServiceName carries Cloud Storage, ServiceCategory Storage, SkuId the SKU and ChargeCategory Usage (FOCUS columns). Which bucket pays for what is in the storage metrics per bucket and class, and in the inventory reports that list objects by class and age.
Levers
Configuration: the first lever is the class per object through a lifecycle rule, by age and by access, because the colder classes cost a fraction per GiB and the rule costs nothing beyond the Class A operation of the transition; the patterns are in storage optimisation. The second is the soft delete window, because a bucket that rewrites its contents pays the rewrite's bytes for seven more days; buckets of temporary data get the window switched off. The third is the operation count: listing a bucket of millions of objects pays Class A operations per 1,000 on every listing, and an application that reads metadata before every object doubles its Class B operations.
Purchase option: none; the tiers of internet egress apply by themselves with volume.
Architecture: a regional bucket next to the compute that reads it, because egress within the region is free and across regions pays per GiB; a dual-region only where two regions actually read; Autoclass for buckets whose access nobody can predict; and the data that leaves Google Cloud behind a CDN, where the cache egress is cheaper than the bucket's (Cloud CDN and Cloud Armor). What traffic costs is in network pricing.
Where it goes wrong
- The Nearline bucket is read every day. Retrieval per GiB and Class B operations exceed the storage saving. The retrieval SKU beside the storage SKU shows it; the objects belong in Standard.
- The Archive object left after a month. Early deletion pays the remaining 335 days.
Early Deleterows are the clue; Archive only for data that stays a year. - Soft delete doubles the daily rewrite. A bucket that replaces 1 TiB a day pays 7 TiB of soft-deleted bytes on top. The bucket's soft delete policy is the lever.
- The listing is the bill. A job that lists a bucket of ten million objects every hour pays 10,000 Class A operations per listing. Inventory reports replace the listing.
- Autoclass on small files. Objects under 128 KiB are not managed and the fee counts the rest; a bucket of thumbnails pays the fee and gains nothing.
- The dual-region pays twice. Storage is billed to both regions and replication per GiB; a backup bucket nobody reads in the second region pays double for nothing.
- Versioning without a lifecycle rule. Every overwrite keeps the old version billable forever. A rule that deletes noncurrent versions after days ends it.
Example
List prices in USD for Frankfurt (europe-west3), read on 2026-10-21 from the Cloud Storage pricing page, per GiB-hour and converted to a month of 730 hours: Standard USD 0.000031507 per GiB-hour, USD 0.023 per GiB-month; Nearline USD 0.000017808, USD 0.013; Coldline USD 0.000008219, USD 0.006; Archive USD 0.000003425, USD 0.0025. Class A operations per 1,000: Standard USD 0.005, Nearline USD 0.01, Coldline USD 0.02, Archive USD 0.05; Class B per 1,000: Standard USD 0.0004, Nearline USD 0.001, Coldline USD 0.01, Archive USD 0.05; retrieval per GiB: Nearline USD 0.01, Coldline USD 0.02, Archive USD 0.05; Autoclass USD 0.0025 per 1,000 objects per 30 days; internet egress to most destinations USD 0.12 per GiB for the first 10 TiB a month, USD 0.11 to 150 TiB, USD 0.08 beyond.
A bucket of 10 TiB (10,240 GiB) costs USD 235.52 a month in Standard, USD 133.12 in Nearline, USD 61.44 in Coldline and USD 25.60 in Archive. The same bucket read in full once a month pays in Nearline USD 102.40 of retrieval on top, USD 235.52 in all, the Standard price; read once a week, USD 409.60 of retrieval, almost twice Standard. A bucket that rewrites 500 GiB a day with soft delete of seven days keeps 3,500 GiB of soft-deleted data billable, USD 80.50 a month beside the live data's USD 11.50. A job that lists one million objects hourly pays 1,000 × 24 × 30 × 0.005 = USD 3,600 a month of Class A operations, more than 150 TiB of Standard storage. Downloading 20 TiB a month to the internet costs 10,240 × 0.12 plus 10,240 × 0.11 = USD 2,355.20. The figures show that the class decides the storage line by a factor of ten, that reads and listings can exceed storage on an active bucket, and that egress is the line that grows with the users rather than the data.
Sources
Read on 2026-10-21. Prices: Cloud Storage pricing. Service: storage classes, Autoclass, soft delete. Billing data: structure of the standard Cloud Billing data export.