S3-compatible storage: what the label covers and what it costs

"S3-compatible" is the most useful and most abused label in storage. The S3 API is the lingua franca - existing tooling works unchanged - so everything wants the label. But compatibility is a spectrum: full API surface with versioning, multipart and lifecycle policies at one end; gateway-only surfaces that silently no-op on advanced features at the other. The ranking records what each surface actually covers; this page gives you the test that tells them apart. The median entry price across the population is $7.99, and the honest test is one afternoon - which is why a compatibility failure discovered after a migration, not before, is this category's most expensive mistake.

The spectrum, from genuine to thin

Three honest tiers [1] - and the tier boundaries are exactly where the durability story diverges too, since several providers in this population publish no durability nines or SLA at all. That is not a decentralized-only problem: the benchmark [5] publishes theirs, and everything else is measured against it. Compatibility and accountability travel together on this ladder:

1. Full API surface - versioning, multipart, lifecycle, IAM-style policies. The genuine article, whether centralized (B2, Wasabi, AWS) or fronting real verifiable networks (filebase's documented model) [1].
2. Core object operations - get/put/list works; advanced features silently no-op. This is where client code breaks in production, three months after the happy-path test.
3. Gateway-only - the S3 shape over token mechanics; rate limits and consistency semantics differ under load. Fine for simple object stores, wrong for anything stateful.

Live from our database - caveats and sources in each linked review.

The three API edges that decide compatibility

Test these with your production client library - the failures are all at the edges [1]:

• Multipart uploads. Large objects die on incompatible multipart semantics - and backups hit multipart constantly. Upload a real 2 GB object in the trial; if it breaks, the label was the marketing.
• List pagination. 1000-key pages, continuation tokens, and prefix behavior differ. A listing bug shows up as silent data loss in cleanup scripts.
• Consistency under load. Read-after-write guarantees vary from strong to eventual. Every provider passes the single-threaded test; run your real traffic shape.

The honest compatibility test is one afternoon with your actual client library at production shapes - cheaper than the migration you avoid. Run it before signing anything: 5 providers cover the population, and every one of them passes the curl test. Only the edges sort them.

Price it: put, keep, and the exit

The S3 API makes providers look interchangeable; the three-layer cost model separates them [2]. Price your real read/write ratio across put, keep and get - the storage cost tool prices the round trip, and the storage hub covers the decentralized backends behind some of these surfaces. An afternoon of API testing plus one priced round trip is the entire due-diligence loop this category needs - the free-trial hub lists the no-card windows to run it in.

Egress: the fee the S3 surface hides

The S3 API makes providers look identical; the egress bill makes them different [2][4][5]:

• AWS S3 - the benchmark: best-documented API, real SLA, and the egress fees that created this entire alternative market [5].
• Backblaze B2 - free egress up to 3x stored volume; genuinely cheap at the $6.95/TB tier [3].
• Wasabi - no egress or API fees with a real 99.9% SLA, held together by the 90-day minimum retention and a 1 TB floor - honest terms that rule out churny workloads [4].
• filebase - free object egress, fronting decentralized backends - with the honest disclosure that the "decentralized" layer is IPFS pinning on provider-controlled infrastructure [1].

Price the exit first: the storage cost tool prices the full round trip at your read ratio, and read-heavy workloads reorder this whole table. The 90-day retention on the no-egress plans deserves its own line in that arithmetic: churny workloads pay for deleted data, archive workloads never notice [4].

Durability: what is actually on offer

The honest disclosure in this population: 80% of the rows carry facts we verified ourselves, and several of them publish no durability SLA at all - the label promises API compatibility, not durability. The durability mechanisms behind each surface are documented in the storage hub; the short version here: erasure coding with audits, collateral-slashed storage networks, and endowments are three different bets that all speak S3.

5 providers make this population - 40% of them without mandatory identity checks, 80% of the rows carrying facts we verified. The spectrum claim is testable on all five rows in one afternoon; the free-trial hub lists which ones give you a no-card window to run it in.