Why an S3 GB-month price is not your archive bill

The GB-month price is only one line

A low storage rate does not price an archive's complete life. AWS separates storage, requests, retrieval, and data transfer; optional services add other charges. Its headline “no minimum charge” does not cancel individual storage classes' minimum-duration rules.[1] This guide examines ordinary S3 objects, not S3 Tables, Vectors, or the separate Glacier vault API, and is not a provider ranking.

The demand is concrete: a public AWS re:Post question asks whether moving one million files creates one million transition requests. Its accepted answer confirms per-object counting.[6] That historical discussion identifies a budgeting problem, not a current price quotation.

Keep each class's minimum separate

• S3 Standard-IA and S3 One Zone-IA: 30-day minimum storage duration and 128 KB minimum billable object size.[1]
• S3 Glacier Instant Retrieval: 90 days and 128 KB minimum billable size; objects remain available for millisecond access.[2]
• S3 Glacier Flexible Retrieval: 90 days; instead of the same 128 KB billing floor, each archived object adds 40 KB of metadata.[1]
• S3 Glacier Deep Archive: 180 days and the same 40 KB metadata structure, not Instant Retrieval's 128 KB billing floor.[1]

For Flexible Retrieval and Deep Archive, 8 KB of that metadata is billed at Standard rates and 32 KB at the archive class's rate. Applying the cheapest archive rate to all 40 KB understates cost.[1] Deleting, overwriting, or transitioning out before a class's minimum expires adds a prorated charge for the remaining period; it does not charge that entire minimum again on top of elapsed storage.[1]

Three different small-object rules

A billing floor, archive metadata overhead, and transition eligibility are different mechanisms. Current Lifecycle defaults prevent objects below 128 KB from transitioning to any class. Configurations predating September 2024 retain the earlier default until modified; explicit size filters can permit smaller objects.[3] Thus a small-object archive estimate must first explain how those objects actually enter that class.

Count objects and inspect their size distribution rather than relying on total capacity. AWS documents one transition request per archived object and suggests considering aggregation to reduce overhead.[3] Aggregation is a trade-off: test whether recovering one small record forces your application to retrieve and unpack a much larger bundle.

A worked bill with explicitly invented rates

All following dollar rates are teaching assumptions, not verified AWS prices. Model one bucket in US East (N. Virginia), USD before tax, a simplified 30-day month, and binary units: 1 GB = 1,073,741,824 bytes, matching S3's storage convention.[1] Here 1 KB means 1,024 bytes. Exclude discounts, optional services, replication, and versioning. Assume the account's applicable internet-transfer allowance is already exhausted.

Upload 1,048,576 objects of 16 KB directly to Standard-IA, then permanently delete them after ten days. Read 65,536 objects once before deletion: 1 GB of payload. Assume storage $0.01/GB-month, PUT $0.01/1,000, GET $0.001/1,000, retrieval $0.02/GB, and chargeable internet egress $0.08/GB.

• Payload: 16 GB. Billable storage after the 128 KB floor: 128 GB.
• Storage: 128 × $0.01 = $1.28 across the minimum period, comprising approximately $0.4267 elapsed storage plus $0.8533 remaining-duration charge.
• Upload requests: 1,048,576 ÷ 1,000 × $0.01 = $10.48576.
• Read requests: 65,536 ÷ 1,000 × $0.001 = $0.065536.
• Retrieval: 1 × $0.02 = $0.02. Egress: 1 × $0.08 = $0.08.
• Total modeled charge: $11.931296, or $11.93. Payload-only storage for ten days would misleadingly suggest 16 × $0.01 × 10/30 = $0.0533.

For the same objects in Deep Archive, a separate steady-state illustration assumes $0.001/GB-month archive storage and $0.02 Standard storage. Metadata makes the monthly storage component (16 + 32) × $0.001 + 8 × $0.02 = $0.208, rather than $0.016. This is not a complete bill: ingestion, retrieval, and the 180-day commitment remain outside that monthly illustration.

Retrieving data is not the same as transferring it

Standard-IA, One Zone-IA, and Instant Retrieval can incur retrieval charges on reads. Flexible Retrieval and Deep Archive require restoration; during the chosen availability window, the archive and temporary Standard-rate copy both incur storage charges.[1] Budget request counts, retrieved bytes, temporary-copy days, subsequent reads, and the destination separately.

Flexible Retrieval's Bulk retrieval requests and data retrieval are free; that does not make the entire recovery free or establish the same rule for Deep Archive. Internet egress has an allowance aggregated across services and regions, with exclusions, while specified same-region transfers are exempt.[1] Neither “all downloads cost money” nor “free retrieval means free download” is a safe budgeting rule.

A decision checklist before changing storage class

• Short retention: compare expected deletion age with the destination's minimum; include the unexpired balance.
• Many tiny objects: calculate size floors or split-rate metadata, then per-object ingestion or transition charges.
• Urgent recovery: check access latency before comparing storage rates; include a realistic restore drill.
• Large external restore: model requests, retrieval, temporary copies, and chargeable egress independently.
• Billing review: reconcile actual bytes, object counts, operations, and destination with each invoice line, rather than dividing the total by stored GB.

Official pages were read on September 22, 2026. Their regional rate tables were not independently verified as a complete current price set, so no numerical rate above is presented as an AWS quote.