I'm currently working with Azure Government, but the GPT-4o experience has been disappointing. Has anyone used both Azure Government and AWS GovCloud for AI workloads? Bedrock looks promising, but I'm wondering how it compares with standard consumer AI tools in terms of model availability, reliability, latency, and overall performance.
2 Answers
AWS GovCloud and Bedrock have improved noticeably. Model availability is broader now, including newer models, and capacity problems that were common a few months ago seem to have eased. The experience can still vary by region, though.
The exact model lineup depends on the region and the current approval and capacity situation. Sonnet 4.5 became much more usable after the initial availability issues, but you should verify which models are enabled in your specific GovCloud region before committing to a production design.
Do you know whether users can access Sonnet 5 in AWS GovCloud, or is that still dependent on regional availability and approval?
Bedrock has been fairly solid, but availability can differ significantly between GovCloud regions. The console generally works well, while API access may be less predictable when new capacity is being brought online. Latency for Sonnet 4.5 was rough for a while but has been much more stable recently, at least in US Gov-West.