S3 Storage Class Pricing: When Each Tier Actually Saves (2026)

Prices below are 2026 US East (N. Virginia) list prices for planning purposes. Exact rates vary by region and change over time — always confirm against the AWS S3 pricing page before budgeting.

Every class below Standard trades a lower storage rate for one or more of these charges. Skip any of them in your mental model and the "cheap" tier can end up costing more than Standard.

Retrieval fees

Standard and Intelligent-Tiering (frequent/infrequent tiers) charge nothing to read data back. Every IA and Glacier class charges per GB retrieved, on top of the storage fee already paid. For Standard-IA that's $0.01/GB — roughly 80% of a full month of Standard storage, paid every time the object is read.

Minimum storage duration

IA classes bill a 30-day minimum; Glacier Flexible Retrieval and Instant Retrieval bill 90 days; Deep Archive bills 180 days. Delete or overwrite an object before that window closes and AWS still charges for the full minimum period (AWS S3 Pricing, 2026 ). Short-lived objects placed in a Glacier tier by mistake can cost more than if they'd never left Standard.

Per-request costs

GET requests get more expensive moving down the tiers — $0.0004/1,000 on Standard versus $0.001/1,000 on IA and up to $0.05/1,000 for a standard Glacier retrieval job. For workloads with millions of small reads, request costs can dwarf the storage savings entirely.

Small-object overhead

IA and Glacier classes bill a minimum object size for storage purposes — 128 KB for Standard-IA and One Zone-IA. A 10 KB object stored in Standard-IA is billed as if it were 128 KB. Buckets full of small files (logs, thumbnails, JSON blobs) can end up paying for storage they don't actually have.

Assume a 100 GB dataset accessed roughly 10 times per month, each read pulling the full 100 GB (a generous, easy-to-follow assumption — real workloads usually read a subset).

Standard-IA looks 46% cheaper than Standard on storage alone, but the retrieval fees make it nearly 5x more expensive in total once real access is factored in. Glacier Instant Retrieval, despite the lowest storage rate of the three, is over 13x more expensive here. This is the crossover: the "discount" tiers only discount storage, and retrieval fees scale with every read.

Now assume 500 GB of compliance/backup data, held for 3 years, read once total (a restore for an audit or investigation) in that period.

Flip the access pattern to true archive and the ranking inverts completely: Deep Archive is roughly 15x cheaper than Standard over three years, because the storage discount compounds every month while the retrieval fee is paid exactly once. This is the same data volume as the first example's conclusion in reverse — the storage class that wins is entirely a function of how often the data gets read, not how "important" or "cold" it feels.

Intelligent-Tiering monitors access patterns per object and automatically moves each one between a frequent-access tier and infrequent-access tier — and optionally into Archive Instant Access, Archive Access, and Deep Archive Access tiers for objects untouched for 90+, 180+ or configurable periods — with no retrieval fees for the frequent and infrequent tiers (AWS S3 Intelligent-Tiering, 2026 ). The cost is a small monitoring fee — $0.0025 per 1,000 objects per month — plus retrieval fees only if objects auto-move into the optional archive tiers and get restored.

Intelligent-Tiering is the right default when:

• Access patterns are unknown, mixed, or unpredictable across the bucket
• Objects are large enough that the per-1,000-object monitoring fee is negligible relative to storage cost (it's a bad fit for buckets with millions of tiny objects)
• You want to avoid the risk of manually picking the wrong tier and eating retrieval fees or early-deletion charges
• The workload has no compliance requirement forcing a specific archive tier

Because there are no retrieval fees on the frequent/infrequent tiers, Intelligent-Tiering effectively removes the crossover risk shown in the worked examples above — you get Standard-IA-like storage savings for cold objects without the retrieval-fee penalty if access resumes unexpectedly.

Manual lifecycle rules — transitioning objects to a specific class after N days — beat Intelligent-Tiering when the access pattern is well understood and predictable:

• Known, uniform access decay — e.g., logs that are hot for 30 days then never read again. A lifecycle rule to Glacier Deep Archive at day 30 skips the monitoring fee and the gradual auto-tiering delay entirely.
• Compliance-mandated retention — regulations often specify an exact class or retrieval SLA; lifecycle rules guarantee the object lands exactly where required rather than relying on access-based automation.
• Very large object counts with predictable lifecycle — the $0.0025/1,000-object monitoring fee adds up at billions of objects; a scheduled lifecycle transition has no per-object monitoring cost.
• Write-once-read-never data (backups, audit logs) — skip the frequent/infrequent tiers altogether and go straight to Glacier via lifecycle rule, since Intelligent-Tiering's auto-tiering delay (30+ days before first archive transition) means paying Standard-IA-equivalent rates during the wait.

The general pattern: Intelligent-Tiering wins under uncertainty, lifecycle policies win when you can state the access pattern in one sentence and are confident it won't change.

Three questions determine the right S3 storage class, in order of importance:

1. How often is the object read? More than monthly → Standard. A few times a year → Standard-IA or One Zone-IA. Less than once a year, or purely for compliance/disaster recovery → Glacier Flexible Retrieval or Deep Archive. Unknown or mixed → Intelligent-Tiering.
2. How big is the average object? Below roughly 128 KB, IA and Glacier classes lose most or all of their storage discount to the minimum billable size and per-request fees — keep small objects on Standard or Intelligent-Tiering regardless of access frequency.
3. How long will it be retained? If retention is shorter than the class's minimum storage duration (30/90/180 days), the early-deletion charge can erase the entire savings — don't archive data you expect to delete or overwrite soon.

One more factor worth checking before committing to a manual tier: can the object tolerate the retrieval latency? Standard-IA and Glacier Instant Retrieval return data in milliseconds; Glacier Flexible Retrieval takes minutes to 12 hours depending on retrieval speed tier; Deep Archive takes 12–48 hours. An archive tier that's otherwise the right price is the wrong choice if a restore needs to complete same-day under an SLA.

For a fast way to model these trade-offs across your own bucket sizes and access patterns without building a spreadsheet, the SpendArk cloud cost calculator lets you compare storage class costs directly. And if S3 costs sit alongside broader data transfer or egress charges, our cloud egress costs guide covers how those stack on top of storage.