internet, information, service, system, room, technology, center, business, rack, hardware, database, computer, data, supercomputer, server, cyber, datacenter, digital, big data, web, 3d, cloud, networking, hosting, security, database, database, supercomputer, supercomputer, datacenter, hosting, hosting, hosting, hosting, hosting
Photo by cookieone on Pixabay

Cost & Limits

Part of AI tool reliability

Reviewing AI availability commitments for business use

Read an AI provider’s availability terms by checking covered traffic, downtime rules, exclusions, credits and your own recovery deadline.

This is an editorial pen name used by the publication, not a claim of an independently verified personal identity. Original editorial credit: Desk Procurement.

undefined

An AI availability commitment is the promise a provider makes about uptime, errors or latency, and one agreement can carry more than one. Review each commitment by naming the covered service, how downtime is counted, the exclusions and the available remedy, then compare it with your own work deadline.

The named example is the Google Cloud service level agreement (SLA) for the Gemini Online Inference API, which defines the Covered Service and sets a Monthly Uptime Percentage Service Level Objective (SLO) for it (cloud.google.com).

Identify the route and business deadline

Record the product, account, plan, model, API method or app feature, relevant region and connected services. Obtain the agreement and order form that would apply to the Australian buyer.

A public service-level page may describe an offering the account has not purchased. Record who can open a support case and what logs the team can supply.

A service bought for business use can still be covered by consumer guarantees when it meets at least one of the conditions the ACCC sets out (accc.gov.au). The ACCC also states that consumers are not covered by the Australian Consumer Law where the business does not officially offer its products and services in Australia.

Ask the task owner when staff must take over. A nightly internal report may tolerate a delay that a customer-facing reply cannot. Set that operational limit even if the provider has not declared an incident.

Read the measurement rule, exclusions and remedy

Check whether the commitment measures uptime, successful requests, server errors or latency. Read the definition of a valid request, the measurement period and any minimum incident duration. Then inspect exclusions for quotas, invalid inputs, preview features, customer configuration and dependent services. A status page may help explain an incident but does not replace the contractual calculation.

Google Cloud Gemini online inference. The current Google Cloud agreement states a 99.5% monthly uptime objective for the covered generateContent and streamGenerateContent methods, or 95% for models designated for shorter availability. Its downtime definition begins by defining “Downtime” as more than a five per cent Error Rate.

The document names a narrow Covered Service: the generateContent and streamGenerateContent methods of the Gemini Online Inference API. For streamGenerateContent under the Provisioned Throughput consumption model, for a Covered Model from a Covered Endpoint, it also sets a Monthly Latency Target Attainment Percentage of 99%. That latency objective does not apply to models designated for shorter availability (cloud.google.com).

If Google does not meet the Service Level Objective (SLO) and the customer meets its obligations under the service level agreement, the customer is eligible to receive Financial Credits, determined on a calendar month basis per Project. The SLA states that this is the customer's sole and exclusive remedy for a failure to meet the SLO. These terms concern the named Google Cloud service, not every Gemini product or request.

One qualification applies to the money. Where the agreement authorises resale or supply of Google Cloud Platform under a partner or reseller program, references to Customer in the SLA mean the partner or reseller, and Financial Credits apply only to impacted partner or reseller orders (cloud.google.com).

Google Cloud Gemini Online Inference API Availability Commitments

  • Service CoveredgenerateContent and streamGenerateContent methods of the Gemini Online Inference API
  • Monthly Uptime SLO99.5% (95% for models designated for shorter availability)
  • Latency Target Attainment99% for streamGenerateContent under Provisioned Throughput model (not applicable to short-availability models)
  • Downtime DefinitionError rate exceeding 5% over a measurement period
  • RemedyFinancial Credits (sole and exclusive remedy) on a calendar month basis per Project
  • ExclusionsQuotas, invalid inputs, preview features, customer configuration, dependent services

Key SLA Metrics for Google Cloud Gemini Online Inference API

Uptime SLO (Standard Models)
99.5%
Uptime SLO (Short-Availability Models)
95%
Latency Target Attainment
99%
Downtime Threshold
>5% Error Rate
Remedy Type
Financial Credits

Decide how work continues

For each candidate, record the covered route, objective, valid-request rule, exclusions, claim process, support channel and task deadline. Mark a missing term as unresolved. Check whether request logs can show the affected time and task without collecting unnecessary business data.

Decide what happens before any contractual claim could help: wait and retry within a safe limit, use an approved alternative or hand the item to a person. Assign an owner to unfinished work. A financial credit may offset a service bill, but it cannot restore a missed business deadline. Revisit the decision after a plan, model, connected service or contract change.

Consumer guarantees can also apply, so keep the contractual remedy and your guarantee position separate. The ACCC notes that getting a repair, replacement or refund from an overseas business can be difficult, and that the help Australian consumer protection agencies can provide is limited (accc.gov.au). Settle the operational fallback before you need either one.

Further reading: Gemini Online Inference API on Gemini Enterprise Agent Platform Service Level Agreement (SLA) | Google Cloud (cloud.google.com).

More from Cost & Limits

Cost & Limits

AI tool costs and usage limits

Assess AI tool costs, staff effort and the usage limits that could interrupt ordinary business work.