S3 Storage Classes Explained: What Each Tier Really Costs in 2026

Every S3 storage class is the same bet: a lower per-GB rate in exchange for slower, costlier reads

Amazon S3 gives you six storage classes you actually assign data to (set Intelligent-Tiering aside — it picks for you). They span a roughly 23x price range, from $0.023 per GB-month at the top down to $0.00099 at the bottom. Each rung down buys a cheaper byte by charging you somewhere else: a slower retrieval, a per-GB retrieval fee, a longer minimum-storage commitment, or one copy of your data instead of three. Here's the full ladder with mid-2026 us-east-1 rates and what each price actually costs you.

The full ladder, with rates and access times

Figures below are us-east-1 (N. Virginia), mid-2026, from the AWS S3 pricing page. Other regions run roughly 10-30% higher, so read these as the floor, not the number on your invoice.

Standard: the default you pay a premium to leave

S3 Standard at $0.023/GB-month is where every object lands unless you tell it otherwise. No retrieval fee, no minimum duration, millisecond access, eleven nines of durability across at least three Availability Zones. The rate steps down above 50 TB ($0.022, then $0.021 past 500 TB), but most buckets never reach the break. This is the right home for data you read often and can't predict — active datasets, assets you serve, anything where a per-request retrieval fee would compound into real money. The entire pricing game below this line is deciding which data does not belong here.

Standard-IA and One Zone-IA: cheaper to store, billed to read

Standard-IA cuts storage nearly in half to $0.0125, then adds a $0.01/GB retrieval fee and a 30-day minimum. The break-even is access frequency. Read a given GB more than about once a month and the retrieval fees swallow the savings. It fits data you keep but rarely touch: a backup set you hope never to restore, logs past their hot window.

One Zone-IA shaves another fifth to $0.01 by holding your data in one Availability Zone instead of three. Same latency, same retrieval fee, one honest catch — lose that zone and the data is gone. AWS still quotes eleven nines within the zone, but the availability SLA is lower and there's no cross-AZ copy to fall back on. Use it for data you can regenerate: secondary copies, derived thumbnails, anything you could re-create from a source you still hold. Never your only copy.

What do the Glacier classes actually trade away?

Glacier is where storage gets genuinely cheap and the access model changes shape. There are three classes, and treating them as one thing is the most common S3 billing mistake people make.

Glacier Instant Retrieval at $0.004/GB keeps the millisecond access of Standard-IA. You pay for that speed with a steep $0.03/GB retrieval fee and a 90-day minimum. It's for archive data you query rarely but need instantly on the rare occasion you do — think medical imaging you might pull on a clinician's deadline.

Glacier Flexible Retrieval at $0.0036 drops the storage rate barely below Instant, but the data goes offline. You issue a restore and wait: expedited in 1-5 minutes and standard in 3-5 hours both carry a fee, while bulk runs free in 5-12 hours. The right tool when hours of latency is fine and you want big restores to cost nothing.

Glacier Deep Archive is the floor at $0.00099/GB-month — about $1.02 to park a terabyte for a month. Retrieval runs 12 to 48 hours and the minimum duration is 180 days. This is tape-replacement storage: the seven-year compliance archive you're legally required to keep and statistically never read.

The retrieval fee is the cost the per-GB number hides

Deep Archive looks unbeatable until you need the data back. Restoring 10 TB carries roughly $200 in retrieval fees on top of the wait — and that's before request charges and the data-transfer-out bill if it leaves AWS. The cheap classes are cheap precisely because AWS is betting you won't read them. For an audit you can schedule a day in advance, that bet is fine. For data a user might ask for at any moment, it's a trap: $1/TB to store, a five-figure surprise to pull it all back. The headline per-GB rate is the easy number; the retrieval fee, the minimum-duration penalty, and the egress charge are where the real cost lives, and none of them show up in the pricing page's first table.

Why a single flat per-TB price exists instead of a ladder

The six-class ladder is a precise instrument and a real source of savings — if your data has clean access patterns and someone owns the lifecycle policies. It's also a forecasting problem. Your bill is a function of how much you read, how fast you needed it, whether objects aged past their minimum duration, and how much left the region. You don't fully control any of those month to month. That uncertainty is baked into the product.

We built beebeeb on the opposite model: one flat rate, €10.99 per TB per month, with retrieval, egress, and access included. No class to pick, no retrieval fee to model, no minimum-duration penalty for deleting early, no surprise when you actually use the data you're paying to store. The tradeoff is honest — a tuned Deep Archive bucket for cold compliance data is cheaper per stored GB than we are, and we won't pretend otherwise. What you get in return is a number you can multiply by your terabytes and trust, on top of full end-to-end encryption: your files are encrypted with AES-256-GCM on your device before they leave it, on every plan, stored in Falkenstein, Germany. Storage scales to 99 TB self-serve at €10.99/extra TB, with a custom quote beyond that.

S3's classes are the right answer when storage is your engineering problem and you have the headcount to run it. A flat per-TB price is the right answer when storage is supposed to stay out of your way. If you'd like to feel what the flat model is like,