I'm currently working in Azure Government, but the GPT-4o experience has been disappointing. Has anyone used both Azure Government and AWS GovCloud for AI workloads? Bedrock sounds promising, but I'm wondering whether its model quality, latency, reliability, and overall user experience are comparable to standard commercial AI tools.
2 Answers
Latency for newer models was noticeably high for a while, but it has been fairly stable recently in at least some western GovCloud regions. I’d still run your own tests with representative prompts, since throughput and response times depend heavily on region, account capacity, model, and request volume.
Bedrock has improved quite a bit. Model availability and capacity were rough a few months ago, especially for newer Claude releases, but things have become more stable and more models are available now. The experience can still vary by GovCloud region and by whether you’re using the console or API.
That matches my experience. Bedrock is generally solid, but model availability can differ significantly between regions. The API may also be inconsistent temporarily when AWS is adding capacity or deploying newer model versions.
Do you know whether users can access newer models such as Sonnet 5 in AWS GovCloud?