S3 GlacierAmazon S3 Glacier storage classes
Three S3 storage classes for archives: Instant Retrieval, read in milliseconds; Flexible Retrieval, restored in minutes to hours; and Deep Archive, the lowest cost, restored within 12 to 48 hours.
Where S3 Glacier takes you5 steps, all open
- I know the problem it solves
- I can picture it
- I know when to choose it
- I can use it in a design
- I can answer an exam question
What & why
Keeping data for years, for lessThe S3 Glacier storage classes keep data you rarely read at the lowest storage costs S3 has. The objects stay in your S3 bucket, designed for the same eleven nines of durability and stored across at least three Availability Zones. What changes is how quickly you can read them back, and what you pay when you do.
Three classes. Instant Retrieval returns an object in milliseconds, like S3 Standard. Flexible Retrieval and Deep Archive are archived: an object must be restored before it can be read, which takes minutes to hours from Flexible Retrieval and up to 48 hours from Deep Archive.
The original Amazon Glacier service, with vaults and an API of its own, no longer accepts new customers. AWS recommends these storage classes instead.
Reach for it when
- backups and records must be kept for years and are rarely read
- regulations require data kept for 7 to 10 years or longer
- images or records are read a few times a year, but must come back in milliseconds when they are
- disaster recovery copies can wait hours to come back
It is the wrong tool when
- the data is read often: that is S3 Standard
- nobody knows how often it will be read: that is S3 Intelligent-Tiering
- the data is kept only a few weeks: a 90 or 180 day minimum is charged anyway
Mental model
A warehouse with three kinds of storageA storage warehouse offers a locker by the entrance that you open any time, a unit at the back that staff fetch within hours, or within minutes for a fee, and deep storage off-site that comes back within two days. The further away a box is kept, the less it costs to keep, and the more it costs to fetch.
Glacier Instant RetrievalGlacier Flexible RetrievalGlacier Deep ArchiveRestoreExpedited, Standard, BulkMinimum storage durationHow it works
In by Lifecycle, out by restore- Get objects into a classUpload straight into a Glacier class, or let an S3 Lifecycle rule move objects there as they age.
- Read Instant Retrieval directlyA normal GET returns the object in milliseconds, with a charge for each GB retrieved.
- Restore the othersAsk for a temporary copy for a number of days, and choose Expedited, Standard or Bulk retrieval.
- Read the copyIt stays readable for the days you asked for. The object itself stays archived, and you pay for both meanwhile.
- Durability
- 11 ninesat least 3 zones, every class
- Minimum duration
- 90 / 90 / 180 daysInstant, Flexible, Deep Archive
- Flexible Retrieval
- 1 to 5 minExpedited; Standard 3 to 5 h; Bulk 5 to 12 h
- Deep Archive
- Within 12 hStandard; Bulk within 48 h
- Per archived object
- 40 KB extraFlexible Retrieval and Deep Archive
- Restore requests
- 1,000 a secondper account
- Storage class
Flexible Retrieval and Deep Archive objects are archived and cannot be read in real time. Instant Retrieval needs no restore at all.
- Restore request
Asks for a temporary copy for a number of days.
- Retrieval tier
Expedited, Standard or Bulk. Deep Archive has no Expedited tier.
- Temporary copy
Readable for the days you asked for, while the object itself stays in its archive class.
Pick a case to follow the restore through each check.
Architecture
Reports that age down the classesThe waterfall only runs downhill. From Flexible Retrieval a Lifecycle rule can move objects only to Deep Archive, and from Deep Archive nowhere. To bring an object back up, restore it and copy it into another class.
Small objects stay put. Objects smaller than 128 KB are not transitioned by default, because every transition is a charged request and small objects save little.
Expire at the end. An expiration action in the same configuration deletes each report when its retention period is over.
Security
Keeping an archive untouchedObject Lock for write-once storage. On a versioned bucket, a retention period or a legal hold stops an object version being overwritten or deleted. In compliance mode nobody can shorten it, not even the root user; in governance mode, users with a special permission can.
Encryption carries over. Encrypted objects stay encrypted through every transition.
A bucket policy cannot stop Lifecycle. Even a policy that denies everything does not prevent a Lifecycle rule transitioning or deleting objects, so guard who can change the rules.
The same access control as the rest of S3. Archived objects are S3 objects, reached only through the S3 console and API, under the bucket's policies and Block Public Access.
Cost
How you are charged, not what it costs todayStorage by the GB-month, lowest in Deep Archive. Each object in Flexible Retrieval or Deep Archive also carries 40 KB of metadata: 8 KB charged at S3 Standard rates and 32 KB at the class's own.
Retrieval by the GB, more for Expedited. Bulk retrieval from Flexible Retrieval is free.
Minimum storage durations. Delete, overwrite or move an object within 90 days, or 180 for Deep Archive, and you pay for the rest of the period. Instant Retrieval also bills every object as at least 128 KB.
Transition requests are charged for each object, which is why small objects are not transitioned by default.
While an object is restored, you pay for both the archived object and its temporary copy.
Prices change, so none are printed here. Check the S3 pricing page on aws.amazon.com.
Comparison
Where rarely read data belongs| Class | First byte | Minimum duration | Availability | Choose it when |
|---|---|---|---|---|
| Glacier Instant Retrieval | Milliseconds | 90 days | 99.9% | Read about once a quarter, but at once |
| Glacier Flexible Retrieval | Minutes to hours, after a restore | 90 days | 99.99% after a restore | Backups and disaster recovery that can wait |
| Glacier Deep Archive | Within 12 or 48 hours, after a restore | 180 days | 99.99% after a restore | Compliance archives kept 7 to 10 years or longer |
| S3 Standard-IA | Milliseconds | 30 days | 99.9% | Read about once a month |
| S3 Intelligent-Tiering | Milliseconds, unless an archive tier is turned on | None | 99.9% | Access patterns unknown or changing |
| Retrieval tier | Flexible Retrieval | Deep Archive |
|---|---|---|
| Expedited | 1 to 5 minutes, for objects under 250 MB | Not available |
| Standard | 3 to 5 hours | Within 12 hours |
| Bulk | 5 to 12 hours, and free | Within 48 hours |
SAA traps
The phrase in the question, and what it meanskept for ten years, almost never read, a request can wait two days, lowest costS3 Glacier Deep Archive
The lowest-cost class, with Standard restores within 12 hours and Bulk within 48.
archive data that must still come back in millisecondsS3 Glacier Instant Retrieval
No restore is needed: a GET returns it at once.
an archived file is needed urgently, within minutesExpedited retrieval from Flexible Retrieval
Typically 1 to 5 minutes for objects under 250 MB. Deep Archive has no Expedited tier.
Expedited retrievals must be accepted even at busy timesProvisioned capacity
Without it, Expedited requests may not be accepted when demand is high.
restore millions of archived objectsS3 Batch Operations
One job restores them all at the rate the account allows.
archived objects deleted after a month were charged for threeThe minimum storage duration
Deleting early is charged for the remaining days: 90 for Instant and Flexible Retrieval, 180 for Deep Archive.
WORM storage that even the root user cannot delete earlyObject Lock in compliance mode
No user can shorten or remove its retention period.
a restored object still shows its Glacier storage classA restore is a temporary copy
The object stays in its archive class. Copy it to change the class.
archiving millions of tiny files costs more than expectedBundle small files into larger objects
Every archived object carries 40 KB of metadata and a charged transition request.
Scenario
Exam style, marked when you answerA hospital must keep scanned records for 10 years. They are almost never read, but when a request for one arrives, the hospital has 72 hours to answer it. The storage must cost as little as possible. Which S3 storage class should it use?
Practice
Fill the gap, the way the challenges workA nightly job writes reports to an S3 bucket. After 90 days they are almost never read, but they must be kept for ten years at the lowest cost, and a request for one can wait up to two days. Where should a Lifecycle rule move them?
Quick revision
The night before the exam- Instant Retrieval
- Milliseconds, no restore;
90days minimum. - Flexible Retrieval
- Restore: Expedited 1 to 5 min, Standard 3 to 5 h, Bulk 5 to 12 h and free.
- Deep Archive
- Lowest cost; Standard within 12 h, Bulk within 48 h;
180days minimum. - Restore
- A temporary copy for the days you ask; the object stays archived.
- Lifecycle
- Only downhill; Deep Archive is one way; under
128 KBstays put by default. - Overhead
40 KBof metadata on every archived object.- WORM
- Object Lock; compliance mode binds even the root user.
- Vaults
- The original Glacier service takes no new customers.
This whole page is free right now.The AWS library is still being written, so every page of it is open to everyone while that lasts. It becomes a paid bundle later; what you read today costs you nothing.
Checked against
Every fact on this page was checked against AWS’s own documentation on 15 Sept 2026. If AWS has changed something since, its page is the one to trust.
- Understanding and managing Amazon S3 storage classes
- Understanding archive retrieval options
- Restoring an archived object
- Transitioning objects using Amazon S3 Lifecycle
- Managing the lifecycle of objects
- Locking objects with Object Lock
- What is Amazon Glacier?
- Amazon S3 Glacier storage classes
- Amazon EBS pricing
- Amazon S3 pricing