Amazon S3 stores every object in a storage class that determines durability, availability, latency, and above all price. Choosing the right class is the single biggest lever on your S3 bill: the same terabyte costs about 23 USD per month in S3 Standard and about 1 USD in Glacier Deep Archive. This tutorial compares the eight current classes, explains the hidden costs (minimum durations, retrieval fees, per-object overhead), shows how to set the class from the CLI, the SDK and Terraform, and how to automate transitions with lifecycle rules.

Prerequisites: an AWS account, the AWS CLI v2 configured, and a test bucket (see S3 bucket creation). Prices below are US East (N. Virginia) list prices as of 2026, rounded, for orientation only; always check the S3 pricing page for your region (Canada Central is roughly 10 % higher).

The comparison table

Storage class	API name	Storage $/GB-month	Retrieval $/GB	First-byte latency	Min. duration	Min. object size	AZs	Typical use
S3 Standard	STANDARD	0.023	none	ms	none	none	≥ 3	Active data, websites, analytics
S3 Intelligent-Tiering	INTELLIGENT_TIERING	0.023 → 0.0125 → 0.004 (auto)	none	ms (opt-in archive tiers: minutes–hours)	none	128 KB monitored	≥ 3	Unknown or changing access patterns
S3 Standard-IA	STANDARD_IA	0.0125	0.01	ms	30 days	128 KB	≥ 3	Backups, older files accessed monthly
S3 One Zone-IA	ONEZONE_IA	0.01	0.01	ms	30 days	128 KB	1	Re-creatable data, secondary copies
S3 Express One Zone	EXPRESS_ONEZONE	0.11	none (request-priced)	single-digit ms	none	none	1 (directory bucket)	ML training, latency-critical workloads
S3 Glacier Instant Retrieval	GLACIER_IR	0.004	0.03	ms	90 days	128 KB	≥ 3	Archives read quarterly (medical images, media)
S3 Glacier Flexible Retrieval	GLACIER	0.0036	0.01–0.03 (bulk free)	1–5 min expedited, 3–5 h standard, 5–12 h bulk	90 days	40 KB overhead	≥ 3	Yearly archives, compliance
S3 Glacier Deep Archive	DEEP_ARCHIVE	0.00099	0.02	12 h standard, 48 h bulk	180 days	40 KB overhead	≥ 3	7–10 year retention, tape replacement

All classes offer 99.999999999 % (11 nines) durability. They differ in availability SLA (99.99 % Standard, 99.9 % IA, 99.5 % One Zone), latency, and the fee structure.

Understanding the hidden costs

Minimum storage duration

Delete or overwrite an object in Standard-IA after 10 days and you are still billed for 30. Deep Archive charges 180 days. Never put short-lived data (temporary exports, CI artifacts) in IA or Glacier classes.

Minimum billable object size

IA classes bill every object as at least 128 KB. A million 4 KB thumbnails cost 128 GB in Standard-IA, more than the 4 GB they would cost in Standard. Glacier classes add 40 KB of metadata per object (8 KB in Standard + 32 KB in Glacier). Rule: small objects stay in Standard, or are bundled into archives before transition.

Retrieval and request fees

Reading 1 TB from Standard-IA costs about 10 USD in retrieval fees on top of requests. If you read the data more than once a month, Standard is cheaper. Glacier Flexible and Deep Archive also require a restore request that produces a temporary copy; bulk restores are the cheapest and slowest.

Transition requests

Lifecycle transitions are billed per 1 000 objects (about 0.01 USD to IA, 0.02 USD to Glacier Flexible, 0.05 USD to Deep Archive). Moving 50 million tiny log files to Deep Archive costs 2 500 USD in transitions alone. Aggregate first.

Worked cost example

10 TB of nightly database dumps, each read maybe once a year for audits:

Class	Monthly storage	Yearly storage	One 10 TB audit restore
Standard	≈ 235 USD	≈ 2 820 USD	0
Standard-IA	≈ 128 USD	≈ 1 536 USD	≈ 100 USD
Glacier Instant Retrieval	≈ 41 USD	≈ 492 USD	≈ 300 USD
Glacier Deep Archive	≈ 10 USD	≈ 120 USD	≈ 200 USD (bulk, 48 h)

Deep Archive wins by a wide margin as long as a 12–48 hour wait is acceptable for audits; if auditors need files within minutes, Glacier Instant Retrieval is the compromise.

Setting the storage class

At upload time (CLI)

BUCKET=my-backups-123456789012

# Single object into Standard-IA
aws s3 cp dump-2026-09-10.sql.gz "s3://$BUCKET/db/" --storage-class STANDARD_IA

# Whole folder straight to Deep Archive
aws s3 sync ./archives/ "s3://$BUCKET/archives/" --storage-class DEEP_ARCHIVE

# Check the class of an object
aws s3api head-object --bucket "$BUCKET" --key db/dump-2026-09-10.sql.gz --query StorageClass
aws s3 ls "s3://$BUCKET/db/" --human-readable      # add --summarize for totals

Changing the class of an existing object

S3 has no “move” call; you copy the object onto itself with a new class. For Glacier Flexible/Deep Archive objects you must restore first.

aws s3 cp "s3://$BUCKET/db/dump-2026-09-10.sql.gz" "s3://$BUCKET/db/dump-2026-09-10.sql.gz" \
  --storage-class GLACIER_IR --metadata-directive COPY

# Restore an archived object for 7 days (bulk tier), then poll
aws s3api restore-object --bucket "$BUCKET" --key archives/2019.tar \
  --restore-request '{"Days":7,"GlacierJobParameters":{"Tier":"Bulk"}}'
aws s3api head-object --bucket "$BUCKET" --key archives/2019.tar --query Restore

From Terraform

resource "aws_s3_object" "report" {
  bucket        = aws_s3_bucket.backups.id
  key           = "reports/2026-q3.pdf"
  source        = "reports/2026-q3.pdf"
  storage_class = "STANDARD_IA"
  etag          = filemd5("reports/2026-q3.pdf")
}

Automating with lifecycle rules

Lifecycle rules move objects between classes as they age and expire them at the end. The classic “backup waterfall” below keeps recent dumps hot, cools them after a month, archives after a year and deletes after seven.

{
  "Rules": [
    {
      "ID": "backup-waterfall",
      "Status": "Enabled",
      "Filter": { "And": { "Prefix": "db/", "ObjectSizeGreaterThan": 131072 } },
      "Transitions": [
        { "Days": 30,  "StorageClass": "STANDARD_IA" },
        { "Days": 365, "StorageClass": "DEEP_ARCHIVE" }
      ],
      "Expiration": { "Days": 2555 },
      "AbortIncompleteMultipartUpload": { "DaysAfterInitiation": 7 }
    }
  ]
}

aws s3api put-bucket-lifecycle-configuration --bucket "$BUCKET" --lifecycle-configuration file://lifecycle.json
aws s3api get-bucket-lifecycle-configuration --bucket "$BUCKET"

The ObjectSizeGreaterThan filter (128 KB) avoids paying the IA minimum on small files, and the multipart rule cleans up abandoned uploads that otherwise cost money invisibly. Our S3 lifecycle management tutorial covers versioned buckets and noncurrent versions.

When Intelligent-Tiering is the right default

Intelligent-Tiering watches each object (≥ 128 KB) and moves it automatically: Frequent Access → Infrequent Access after 30 days without reads → Archive Instant Access after 90 days, back to Frequent on the next read, with no retrieval fees. The price is a small monitoring fee (0.0025 USD per 1 000 objects per month). For data lakes, user uploads and anything whose access pattern you cannot predict, it is the safest default. It is a poor fit for millions of tiny objects (monitoring fee dominates) and for data you know is cold from day one (go straight to Glacier).

Decision flow

1. Need single-digit millisecond latency for hot, compute-adjacent data? → Express One Zone.
2. Accessed often or unpredictably? → Standard (known hot) or Intelligent-Tiering (unknown).
3. Accessed about monthly, objects > 128 KB, kept > 30 days? → Standard-IA; if easily re-creatable → One Zone-IA.
4. Accessed quarterly or less, but must be instant when needed? → Glacier Instant Retrieval.
5. Hours of delay acceptable, kept > 90 days? → Glacier Flexible Retrieval.
6. Compliance archive, 12–48 h delay acceptable, kept > 180 days? → Glacier Deep Archive.

Key takeaways

• Durability is identical across classes; you are trading availability, latency and fees for storage price.
• Watch the three traps: minimum duration, 128 KB minimum size, and retrieval/transition fees.
• Intelligent-Tiering is the safe default for unpredictable access; Deep Archive is the tape replacement.
• Automate with lifecycle rules filtered by prefix and object size, and always clean incomplete multipart uploads.

Next tutorial

Next: S3 lifecycle management, then cross-region replication. Official docs: Using Amazon S3 storage classes.