At a glance
| Slab | Swarm (Bee) | Storj | Filecoin | Arweave | Cloud (S3 and similar) | |
|---|---|---|---|---|---|---|
| Token | none | BZZ | STORJ (operator payouts) | FIL | AR | none |
| Blockchain | none | Gnosis Chain | for token payouts | its own | its own | none |
| You pay in | euros, per slab (simulated on the testnet) | BZZ postage stamps | dollars | FIL deals | AR, once | dollars or euros |
| Encrypted before upload | always | optional | by default | up to you | up to you (public by default) | up to you |
| Can payments be linked to content? | no: blind-signed vouchers | yes: every chunk carries its batch’s stamp | the coordinator (satellite) knows which account stores which objects | deals are public on chain | transactions are public | the provider knows both |
| Redundancy | erasure coding: 1.6× today, 1.5× next | replication in a neighbourhood (about 5× measured) | erasure coding | copies per deal | many full replicas | provider-internal |
| Node memory at idle | 23–30 MiB (measured) | about 710 MiB median (measured) | n/a | heavy (sealing hardware) | n/a | n/a |
| Who runs it | independent operators | independent operators | independent operators + coordinators | storage providers | miners | one company |
| Maturity | testnet proof of concept | production | production | production | production | production |
Slab and Bee numbers are measured: Slab on our 20-node testnet, Bee on 1,310 mainnet full nodes (Bee 2.8.3, November 2026). The other columns summarize public documentation and simplify; corrections welcome.
Compared with Swarm
Slab started as a question: what is left of Swarm’s Bee if payments and the chain leave the node? We measured both.
| Bee full node | Slab node | |
|---|---|---|
| Memory at idle | median 706–720 MiB | 23–30 MiB |
| CPU at idle | median 13–23 millicores | 1–8 millicores |
| Bytes stored per byte uploaded | about 5 (neighbourhood size) | 1.6 (10+6), 1.5 with 32+16 |
| Code | chain and payment code larger than all of Slab | about 6× smaller, no chain code |
| Large-file reads | Swarm mainnet | up to 25× faster on our testnet* |
*The testnet sits in one data centre, so the read speed doesn’t prove a protocol advantage yet.
Where Slab is behind:
- Price: not comparable yet. On the testnet the price is a setting in the control plane, paid in simulated euros, so it can be anything. What carries over is cost: a node stores 1.6 bytes per byte uploaded (1.5 next) against about 5 on Swarm, and pays no gas. The real price will come from the market (roadmap phase 7).
- CPU with data. Once it holds data, a Slab node idles at about 0.075 cores, versus 0.029 for a Bee full node with a much larger reserve, because repair checks grow with stored data.
- Scale. Swarm runs thousands of nodes. Slab runs 20 on two hosts.
Compared with cloud storage
Cloud object storage is fast, cheap at scale and run by one company, which can read anything you don’t encrypt and can close your account. Slab encrypts everything on your machine, spreads each file over many operators so no one of them can lose it or hand it over, and the service that sells storage never learns which files are yours. Slab is not as fast or as cheap yet.
What only Slab does (as far as we know)
- Blind-signed payment. Storage is paid with vouchers that the seller signs without seeing them (RFC 9474). Nothing links an account to the slabs its vouchers paid for, not even for us.
- A storage node with no money in it. No wallet keys, RPC endpoints or balances on the node: our build checks forbid them. That is why it fits in about 25 MiB.
- Paid for delivery, provably. Downloads produce signed delivery receipts, and nodes are paid from them, capped per shard so self-dealing doesn’t pay.
- Earnings as storage (planned). An operator can take storage instead of cash, so contributing disk pays for your own backups.