Best LLM Inference Providers

ComparisonChecked Nov 8, 2026

Amazon Bedrock vs Google Cloud Vertex AI

Both put proprietary and open-weight models inside a cloud account with that cloud’s IAM, networking, and billing, and both offer OpenAI-compatible Chat Completions endpoints. They describe data retention and compliance scope differently.

What differs

  • Models and custom weights: Bedrock offers models from many providers plus Custom Model Import. Google offers Gemini and other models, plus self-deployed open models from Model Garden.
  • Data retention: Bedrock sets retention per AWS Region through modes (none, default, aws_review); some models require AWS retention or review and are unavailable under zero retention without approval. Google caches Gemini data by default for up to 24 hours and documents feature-by-feature steps toward zero data retention.
  • Published compliance: AWS says Bedrock is in scope for ISO and SOC, HIPAA eligible, usable in compliance with GDPR, and FedRAMP High in GovCloud (US-West). Google lists Generative AI on Gemini Enterprise Agent Platform in its SOC 2 and ISO 27001 scope and as a HIPAA-covered product.
  • Naming: Google’s docs now call the platform Gemini Enterprise Agent Platform; contracts may use either name.
  • Pricing: both publish per-token pricing by model.

Who each tends to fit

The choice usually follows your existing cloud and which proprietary models you need. Confirm the retention mode and region for every model before production.

Last checked: . No winner declared.

Side by side

FactAmazon BedrockGoogle Cloud Vertex AI (Gemini Enterprise Agent Platform)
StatusActive. Amazon Bedrock is a service and business line of Amazon Web Services (Amazon.com, Inc.).Active. A Google Cloud service and business line of Google (Alphabet Inc.). Google renamed several Vertex AI products under Gemini Enterprise Agent Platform.
OwnershipSubsidiarySubsidiary
Buying modelServerless per-token API, Dedicated or provisioned capacity, Bring your own model, Cloud provider platformServerless per-token API, Dedicated or provisioned capacity, Bring your own model, Cloud provider platform
CategoriesCloud provider platform; Serverless per-token API; Provisioned throughput; Bring-your-own-model (Custom Model Import)Cloud provider platform; Serverless per-token API; Provisioned throughput; Bring-your-own-model (Model Garden self-deploy)
PricingPublished per-token pricing on https://aws.amazon.com/bedrock/pricing/ (by model and Region). AWS states batch inference for select models is priced 50% lower than on-demand.Published per-token pricing on https://cloud.google.com/vertex-ai/generative-ai/pricing (for example, introductory pricing of $0.75 / $3.75 per 1M input / output tokens for Gemini 3.8 Flash through December 31, 2026).
CertificationsSOC (in scope); ISO (in scope); CSA STAR Level 2; HIPAA eligible; GDPR (AWS states customers can use Bedrock in compliance with the GDPR); FedRAMP High (AWS GovCloud US-West)SOC 2 (Generative AI on Gemini Enterprise Agent Platform in scope); ISO 27001 (Generative AI on Gemini Enterprise Agent Platform in scope); HIPAA (Generative AI on Agent Platform listed as a covered product)
LimitationsRetention depends on the model: some models require AWS retention or AWS human review (aws_review mode, up to 30 days) and are unavailable under zero-retention mode unless approved; Data retention settings are per Region and do not propagate to other Regions; Model availability and prices vary by AWS RegionZero data retention needs configuration: caching is on by default, some grounding features store data and cannot be disabled, and ZDR for some partner models requires approval; Google may log prompts for abuse detection under its terms; Product names changed from Vertex AI; check current names in contracts and docs
How to buyaws.amazon.comcloud.google.com
Sources[1] [2] [3] [4] [5] Checked 2026-11-08[1] [2] [3] [4] [5] [6] [7] Checked 2026-11-08

Facts come from each vendor’s profile sources. Run a paid pilot before choosing.