Skip to content
Uptime.
Updated 22h ago

Cloud SLAs: promise vs reality

82 services from 22 vendors with an SLA; 26 used more downtime than their SLA allows in the last year.

82 of 82 services

Promised in the SLAObserved on the vendor's status pagePartial historyShortfall

Rankings

All rankings

Recent SLA changes

Every change
  • Sep 23, 2026OpenAI API Scale TierGPT-5.5 Scale Tier latency SLA listed as '99% > 100 tokens per second' in the 2026-06-11 snapshot; the live page on 2026-09-23 lists '99% > 50 tokens per second'. Exact change date unknown (between those dates).Source
  • Sep 23, 2026Anthropic Claude API Priority TierPriority Tier moved from a purchasable offer ('Get started with Priority Tier ... Higher availability: Target 99.5% uptime', still present in the 2025-11-11 snapshot) to 'no longer available for purchase' for new commitments. The 99.5% target is unchanged. Exact change date unknown; after 2025-11-11, observed 2026-09-23.Source
  • Jul 30, 2026OpenAI API Fast mode (formerly Priority processing)Priority processing renamed Fast mode ('Priority processing was renamed Fast mode on July 30, 2026'); service_tier 'priority' still accepted.Source
  • Jun 1, 2026GitHub Enterprise CloudCredit schedule changed. The June 2021 version (Wayback snapshot 2022-02-03, PDF) gave 10% of fees when quarterly Uptime was <= 99.9% but > 99.0%, and 25% below 99.0%. The current 'Version: June 2026' page gives 5% for 99.5-99.9%, 10% for 99.0-99.5%, 25% below 99.0%, so small misses now earn half the prior credit. The 99.9% quarterly commitment is unchanged. A 2025-12-31 snapshot still linked the 20210630 PDF, so the change landed in the first half of 2026; exact day not published (date set to the version month).Source
  • Feb 24, 2026Snowflake Postgres (HA instances)Appendix A added a Snowflake Postgres HA Feature Level of 99.95% with 3x/7x/total HA-instance credits; not present in the January 9, 2026 version.Source
  • Feb 5, 2026Databricks Data + AI PlatformSupport offerings table updated: the Enhanced tier was dropped from the current table (Business / Production / Mission Critical remain); Enhanced details retained on a separate page for existing commit customers. Page: 'Last update to offerings table February 5, 2026, for earlier version with details on Enhanced Support for existing commit customers'.Source

Summaries of published SLAs; the contract you sign governs. Logos via logo.dev; trademarks belong to their owners.

Questions

Which cloud provider has the best SLA?

Per service, the highest published commitments are 99.99% for multi-zone compute and databases at AWS, Azure and Google Cloud, and 99.999% for a few multi-region databases. Cloudflare promises 100% on its Business and Enterprise plans. See the promised ranking for the full list with sources.

What does 99.9% uptime mean in minutes?

99.9% allows about 43.8 minutes of downtime a month (8.8 hours a year). 99.95% allows 21.9 minutes a month, 99.99% about 4.4 minutes, and 99.999% about 26 seconds.

What do I get when a vendor misses its SLA?

Almost always a service credit: a percentage of that month’s bill for the affected service, usually 10% to 30%, sometimes up to 100%, and only if you claim it within the vendor’s window. Credits are the sole remedy in most SLAs.

Does OpenAI or Anthropic have an SLA?

OpenAI publishes a 99.9% uptime SLA only for its Scale Tier; the standard API has none. Anthropic states a 99.5% uptime target for its Priority Tier, which is not sold to new customers. Enterprise contracts may include their own terms.

How is observed uptime calculated?

From each vendor’s own status page: minutes during which the vendor marked a component of the service as a major outage (or rated the incident critical), overlaps merged, worst region, over the last 90 or 365 days. Partial outages and degraded performance are shown but not counted. It is a proxy for what customers saw, not a measurement of your account.

Weekly: SLA changes from cloud, data and AI vendors, Fridays.