I wrote an S3 deep dive earlier this year. It was one article, and it kept growing questions it couldn't hold: what does this workload cost in Glacier Instant Retrieval versus R2, which aws s3 commands delete things, who actually left S3 and why. So I kept going until it was a site.

S3 Atlas home page: the storage classes stacked as a water column, Express One Zone at the surface and Glacier Deep Archive at the bottom, with $/GB-month for each

Two halves, one source of truth

docs/ has 18 chapters, about 15,000 lines per language. I researched them in Japanese and translated to English, and CI checks that every Japanese chapter has an English counterpart with the same number of headings. Each chapter ends with its references. Anything I could not confirm stays marked unverified in the text; six of those marks are left in the English docs.

data/ is the structured part, and the React app builds pages on it:

Page Data
Storage classes 10 classes as layers of a water column
Cost calculator every S3 class plus 21 other services
Competitors 22 services: price sheet, feature matrix, profiles
Cases 38 public cases, including Dropbox and 37signals leaving S3
Timeline 68 events, 2006-03-14 launch ($0.15/GB-month) to 2026-10-01
Commands 134 CLI, SDK and third-party recipes, 17 flagged destructive
API 309 operations across S3, S3 Control, Tables, Vectors, Files, Outposts

The header toggle switches every page, chapter and dataset to Japanese.

The calculator

Cost calculator with the 10 TB media library preset: Wasabi $82, Hetzner $98, Storj $108, Backblaze B2 $122, Cloudflare R2 $159 at the top, each bar split into storage, requests, retrieval and egress

It ignores free tiers for every provider so the comparison stays even. What it does not ignore is per-object overhead, which is where S3 class choice usually goes wrong. IA and Glacier Instant Retrieval bill anything under 128 KB as 128 KB. Glacier Flexible Retrieval and Deep Archive add 32 KB at the class price plus 8 KB at the Standard price for every object.

I ran the site's own s3ClassCost from src/lib/cost.ts against data/ for 1 TB with 100k PUTs and 100k GETs, changing only the average object size:

1 TB, 1024 KB average object S3 Standard $ 24.09/month S3 Standard-Infrequent Access $ 13.90/month S3 Glacier Instant Retrieval $ 7.10/month S3 Glacier Deep Archive $ 6.27/month 1 TB, 16 KB average object S3 Standard $ 24.09/month S3 Standard-Infrequent Access $ 103.50/month S3 Glacier Instant Retrieval $ 35.77/month S3 Glacier Deep Archive $ 19.86/month

Same terabyte. At 16 KB objects, Standard-IA costs four times what Standard does.

Competitor rows carry their own catch where one exists. Wasabi wins the preset above at $7.99/TB-month with no egress or API fees, and its entry also lists the 90-day minimum, the 1 TB minimum charge, and the fair-use rule that egress stays free only while it is under what you store.

Keeping it from going stale

Prices are us-east-1 list prices, checked against the AWS Price List API and vendor pricing pages as of 2026-10-03, and each data file carries a verifiedDate. freshness.yml runs on the 1st and 15th: it pulls the AWS What's New feed, lists S3 announcements newer than the last timeline entry, flags data files verified more than 120 days ago, and puts both in a tracking issue.

The hands-on labs chapter has 12 labs. Labs 1 to 3, 7, 9 and 12 were run against pgsty/silo (the community fork that kept building the MinIO server after MinIO stopped shipping images). The rest needed real AWS, so only their CLI arguments were checked against a moto server, and the chapter says so.

Locally:

$ node scripts/validate-data.mjs ✓ data (10 files) and docs parity OK $ npx vitest run Test Files 4 passed (4) Tests 43 passed (43)

S3 Atlas is an independent study guide, not affiliated with AWS. Corrections are welcome as issues: https://github.com/0-draft/s3-atlas