Meter
What a durability number promises, and what an SLA actually pays out
Durability and availability are different promises. Every durability number here is a vendor-published design claim that we are republishing, not a result we have independently verified. Availability SLAs usually remedy qualifying downtime with service credits, not recovery of lost backup data.
Put your own numbers through the calculator to see what this meter does to a normal month and to the day you restore.
A durability claim is not an availability SLA
Durability estimates the chance that stored objects survive over time. Availability measures whether the service can answer eligible requests. A vendor can publish eleven nines of designed durability and a 99.9% availability SLA without contradiction because the two percentages measure different failure modes. [1] [2]
We did not test any vendor's annual durability. The figures below are the vendors' own published design claims, carried with their wording and retrieval dates. A durability percentage should not be read as independent validation, a recovery-time promise, or a guarantee that deleted or corrupted customer data will be restored. [5] [6]
An availability SLA is also narrower than an uptime headline. Eligibility, excluded events, claim deadlines, affected-resource calculations, and plan restrictions determine whether a credit is owed. The usual remedy is a credit against service fees, not data reconstruction, operational recovery, or compensation for business loss. [3] [4]
| Published property | What it measures | Usual remedy | Source ref |
|---|---|---|---|
| Durability | A vendor's modeled or designed annual probability of retaining objects | No remedy follows from the percentage unless contract terms separately create one | [1] |
| Availability | Successful access or requests during the SLA measurement period | conditional service credits after a qualifying breach and approved claim | [2] [3] |
| Service credit | A percentage of eligible fees for the affected service or resource | Future bill relief in most checked SLAs, not recovery of the backup | [3] [4] |
| Data loss or corruption | A different failure from service unavailability | Often excluded from the availability calculation or expressly left to the customer's independent backup plan | [5] [6] |
Published claims and contractual remedies differ across 18 vendors
Each provider remains visible. not published means the checked official source did not publish the requested number or remedy; it never means zero. conditional identifies plan, class, region, or contract restrictions. stale marks a previous official term that is not safe to present as current. not comparable preserves a source disagreement instead of choosing a preferred value.
The durability column republishes vendor claims only. It does not convert architecture descriptions into invented percentages. The availability column distinguishes formal SLA commitments from marketing uptime, and the remedy column states what an approved claim buys rather than implying that a credit restores data.
Published claims and contractual remedies differ across 18 vendors18 rows
| Provider | Published durability | Availability SLA | What the SLA remedies | Source ref |
|---|---|---|---|---|
| Amazon S3 | 99.999999999% annual durability, described as 11 nines and designed for S3 Standard [7] | conditional by class: 99.9% for Standard, Express One Zone, Glacier Flexible Retrieval, and Glacier Deep Archive; 99.0% for Intelligent-Tiering, Standard-IA, One Zone-IA, and Glacier Instant Retrieval [3] | 10%, 25%, or 100% service credit by breach band, generally against future S3 payments; sole and exclusive availability remedy, not data recovery [3] | [7] [3] |
| Azure Blob Storage | At least 99.999999999% annual durability for LRS, 99.9999999999% for ZRS, and 99.99999999999999% for GRS or GZRS [8] | conditional by tier, redundancy, and request direction: 99.9% for hot-tier writes and most hot reads, 99.99% for eligible RA-GRS or RA-GZRS hot reads, 99.0% for cool, cold, or archive writes, and 99.9% for eligible cool, cold, or archive reads [4] | 10% or 25% service credit by applicable threshold. Credits are the sole and exclusive performance or availability remedy and do not compensate other loss [4] | [8] [4] |
| Google Cloud Storage | At least 99.999999999% annual durability for all storage classes and location types [9] | conditional by class and location: commitments span 99.0%, 99.9%, and 99.95%, with separately listed regional exceptions [10] | 10%, 25%, or 50% financial credit against future bills after an eligible claim; no data-recovery remedy is stated [10] | [9] [10] |
| Cloudflare R2 | 99.999999999% annual durability, described as 11 nines and designed rather than independently measured here [1] | conditional: 99.9% monthly uptime for Enterprise customers covered by the R2 SLA [2] | 10%, 25%, or 50% service credit against future R2 usage; sole and exclusive remedy for SLA failure, not restoration of data [2] | [1] [2] |
| Backblaze B2 | 99.999999999% annual durability calculated for the published vault architecture [11] | 99.9% monthly service-level objective for B2 Cloud Storage [12] | 5% credit from 99.0% to below 99.9%, or 10% below 99.0%, applied to the next payment; sole and exclusive remedy, not data recovery [12] | [11] [12] |
| Wasabi | 99.999999999% annual durability, described as designed 11 nines [13] | Published system availability is 99.9% for a single region and 99.99% across two regions. The checked current legal SLA page did not expose its operative body, so the current contractual schedule is not published from that page [14] [15] | stale: the current page links a February 2024 reference-only PDF with 10% and 25% credits capped at 50% and applied to future service. Current remedy terms are not published in the checked live body, so the old schedule is not presented as current [15] [16] | [13] [14] [15] [16] |
| IDrive e2 | 99.999999999% durability, described as eleven nines with three copies of data [17] | At least 99.9% monthly uptime for paid IDrive services including IDrive e2 [18] | 5% credit from 99.0% to below 99.9%, or 10% below 99.0%, normally against future invoices with a discretionary refund option; sole and exclusive remedy, not data recovery [18] | [17] [18] |
| DigitalOcean Spaces | not published: official feature documentation describes Ceph redundancy and says Cold has the same durability as Standard, but gives no numeric durability figure [19] | conditional by product: 99.9% for Spaces Standard and 99.5% for Spaces Cold Storage [5] [20] | Standard credits are 10%, 25%, or 100%; Cold credits are 5%, 10%, 25%, or 100%. Both are future-use credits, data loss is excluded, and the service is not intended as the sole backup [5] [20] | [19] [5] [20] |
| Scaleway Object Storage | 99.999999999% durability for Multi-AZ Object Storage [21] | conditional by deployment: 99.9% for Standard Multi-AZ and 99.0% for One Zone Infrequent Access or Glacier [22] | 10%, 25%, or 50% compensation voucher, capped at 50% of the affected month's fees; the voucher is the all-inclusive remedy and does not recover data [22] | [21] [22] |
| OVHcloud Object Storage | not comparable: the same official sentence prints 99.99999999999% and labels it 11 nines. Those forms are mathematically inconsistent, so both are reported and neither is silently selected [23] | conditional by deployment: 99.9% in 1-AZ and Local Zones, and 99.99% in 3-AZ regions [23] | Service compensation rather than recovery: 0.5% of affected monthly consumption per minute of access failure, capped at 100%; missing the separate 100% monthly data-resilience commitment compensates 100% of affected consumption [24] | [23] [24] |
| Hetzner Object Storage | not published: Hetzner documents erasure coding that tolerates up to three storage-server failures, but no numeric durability percentage [25] | not published: no Object Storage availability SLA was found in the checked official Object Storage documentation [25] | not published: no Object Storage service-credit schedule was found; the documentation says Object Storage should be only part of a backup strategy [25] | [25] |
| Akamai Cloud Object Storage | 99.999999999% annual durability for E2 and E3 endpoints, with documented limits for disasters, fire, and human error [26] | 99.9% monthly uptime for Object Storage under the current Non-Compute Service Level Addendum [27] | Pro-rata service credit for an affected monthly billing period after an eligible claim; the SLA provides billing credit, not reconstruction of objects [27] | [26] [27] |
| Vultr Object Storage | not published: current product documentation calls the service durable and highly available but does not publish a numeric durability percentage [28] | not published: no numeric Object Storage availability SLA was found in the checked current Object Storage documentation [28] [29] | not published: no Object Storage service-credit remedy was found in those sources [28] [29] | [28] [29] |
| Storj | 99.999999999% durability, described as 11 nines for all published pricing tiers [30] | 99.95% monthly availability for Storage Services; Edge Services have a separate 99.9% commitment [31] | 5% credit below 99.95% and above 99.0%, or 10% below 99.0%, capped at 10% and applied to future fees; sole and exclusive remedy, while retrievability is not guaranteed [31] | [30] [31] |
| Oracle OCI Object Storage | 99.999999999% annual durability, described as eleven nines and designed for OCI Object Storage [32] | not comparable across current official sources: the tier guide lists 99.9% for Standard, 99.0% for Infrequent Access, and no numeric Archive value, while the March 2026 pillar covers Object Storage and Archive Storage SKUs with a 99.9% API-uptime commitment [33] [34] | 10%, 25%, or 100% service credit for eligible Object Storage availability breaches; a billing credit does not recover deleted or overwritten data [34] [32] | [32] [33] [34] |
| Tigris | not published: Tigris says it offers the same durability guarantees as large clouds but publishes no numeric durability figure on the checked page [35] | not published as a public SLA: the site markets 99.99% uptime, while pricing says larger workloads can customize a service contract [35] [36] | not published: no public credit schedule was found, and a customized enterprise contract is conditional on negotiation [36] | [35] [36] |
| Filebase | not published: Filebase describes its object storage as durable but does not publish a numeric durability percentage on the checked product page [37] | conditional: public self-service plans have no published numeric SLA; Enterprise advertises a dedicated SLA without public terms [38] | not published: the public pricing page gives no credit percentage or data-recovery remedy for the Enterprise SLA [38] | [37] [38] |
| Telnyx Cloud Storage | not published: the product page says data is duplicated across multiple datacenters with no single point of failure, but gives no numeric durability percentage [39] | not published: the current terms provide Storage Services as is and as available rather than publishing an availability commitment [6] | does not apply under the public terms: Telnyx disclaims storage warranties and responsibility for loss, misuse, deletion, or failure to store, and assigns backup responsibility to the customer [6] | [39] [6] |
Egress still decides the restore bill
Durability and SLA complete the risk picture, but egress remains the cost variable that can dominate a backup restore. The rules below are carried from the 61-offer, 18-vendor provider registry and its cited vendor sources; they are not recalculated from storage rates. [40] [41] [42]
A restore is read-only in this model. Its possible meters are retrieval, API requests, temporary-copy storage, and egress. It never triggers early deletion, which applies only when an object is deleted, overwritten, moved, or transitioned before its minimum. [43] [44] [45]
| Provider or action | Restore treatment | State | Source ref |
|---|---|---|---|
| Backblaze B2 | Internet egress is free up to 3 times average monthly storage, then $0.01/GB | conditional | [40] |
| Cloudflare R2 | Internet egress has no charge, but Class A and Class B operations are billed, so a restore is not automatically a zero-cost event | not charged for egress; conditional for requests | [41] |
| Wasabi | No egress fee while monthly downloads stay within the fair-use policy tied to active stored volume. An excess per-GB rate is not published; policy enforcement must not be converted to $0 overage | conditional; excess rate not published | [42] |
| Read-only restore | Retrieval, requests, temporary-copy storage, and egress can apply; no delete, overwrite, move, or transition occurs | Early deletion: does not apply | [43] [44] [45] |
Sources
Every figure above comes from one of these pages, on the date shown. Each numbered reference in the text links to its entry here. Vendors change prices; if a date looks old, check the source. Three vendors publish a machine-readable feed and the rest are re-checked by hand, which is why the dates are not uniform.
The source ledger45 sources
- 1Cloudflare developer docs: Durabilitydevelopers.cloudflare.com · checked 2026-07-26
- 2Cloudflare: R2 service level agreementwww.cloudflare.com · checked 2026-07-26
- 3Amazon Web Services: Slaaws.amazon.com · checked 2026-07-26
- 4Microsoftwww.microsoft.com · checked 2026-07-26
- 5DigitalOcean: Spaceswww.digitalocean.com · checked 2026-07-26
- 6Telnyx: Terms and conditions of servicetelnyx.com · checked 2026-07-26
- 7Amazon Web Services: Faqsaws.amazon.com · checked 2026-07-26
- 8Microsoft Learn: Storage redundancylearn.microsoft.com · checked 2026-07-26
- 9Google Cloud: Availability durabilitycloud.google.com · checked 2026-07-26
- 10Google Cloud: Slacloud.google.com · checked 2026-07-26
- 11Backblaze help centre: B2 Resiliency Durability and Availabilityhelp.backblaze.com · checked 2026-07-26
- 12Backblaze: Slawww.backblaze.com · checked 2026-07-26
- 13Wasabi documentation: Durability security reliabilitydocs.wasabi.com · checked 2026-07-26
- 14Wasabi documentation: What is wasabis system availabilitydocs.wasabi.com · checked 2026-07-26
- 15Wasabi: Slawasabi.com · checked 2026-07-26
- 16Brandfolder: SLA 21Feb2024 REFERENCE ONLYcdn.brandfolder.io · checked 2026-07-26
- 17IDrivewww.idrive.com · checked 2026-07-26
- 18IDrive: Terms slawww.idrive.com · checked 2026-07-26
- 19DigitalOcean documentation: Featuresdocs.digitalocean.com · checked 2026-07-26
- 20DigitalOcean: Spaces cold storagewww.digitalocean.com · checked 2026-07-26
- 21Scaleway: Object storagewww.scaleway.com · checked 2026-07-26
- 22Scaleway: Slawww.scaleway.com · checked 2026-07-26
- 23OVHcloud: Classeswww.ovhcloud.com · checked 2026-07-26
- 24OVHcloud: Public cloud faqwww.ovhcloud.com · checked 2026-07-26
- 25Hetzner documentation: Generaldocs.hetzner.com · checked 2026-07-26
- 26Akamai techdocs: Data durabilitytechdocs.akamai.com · checked 2026-07-26
- 27Akamai: Sla non computewww.akamai.com · checked 2026-07-26
- 28Vultr documentation: Object storagedocs.vultr.com · checked 2026-07-26
- 29Vultr documentation: Limitsdocs.vultr.com · checked 2026-07-26
- 30Storj: Pricingwww.storj.io · checked 2026-07-26
- 31Storj: Terms of servicewww.storj.io · checked 2026-07-26
- 32Oracle: Object storage faqwww.oracle.com · checked 2026-07-26
- 33Oracle documentation: Understandingstoragetiersdocs.oracle.com · checked 2026-07-26
- 34Oraclewww.oracle.com · checked 2026-07-26
- 35Tigriswww.tigrisdata.com · checked 2026-07-26
- 36Tigris: Pricingwww.tigrisdata.com · checked 2026-07-26
- 37Filebase: Object storagefilebase.com · checked 2026-07-26
- 38Filebase: Pricingfilebase.com · checked 2026-07-26
- 39Telnyx: Cloud storagetelnyx.com · checked 2026-07-26
- 40Backblaze: Transaction pricingwww.backblaze.com · checked 2026-07-26
- 41Cloudflare developer docs: R2 pricingdevelopers.cloudflare.com · checked 2026-07-26
- 42Wasabi: Faqwasabi.com · checked 2026-07-26
- 43AWS documentation: Restoring objectsdocs.aws.amazon.com · checked 2026-07-26
- 44Amazon Web Services: S3 pricingaws.amazon.com · checked 2026-07-26
- 45Google Cloud: Storage pricingcloud.google.com · checked 2026-07-26