Lower your AI inference costs with new Bedrock model options

AWS just added new model choices on Amazon Bedrock. GPT-6.1 Sol is now generally available, and AWS says it delivers strong performance at about one-fifth the cost of GPT-6 Astra.

That cost difference makes this one especially interesting for teams running agentic workloads at scale. If you’re watching inference spend, having a lower-cost option with solid performance gives you more room to optimize without giving up speed.

Get better cloud cost visibility and stronger allocation controls

AWS Billing and Cost Management added a new billing context API. The new ListBillingViewSegments API returns billing context for a time period, including billing hierarchy and rate settings.

That’s useful when you’re trying to understand how accounts appear in chargeback and cost-allocation workflows. Better billing context can make it easier to explain costs, segment usage, and keep reporting aligned with your internal structure.

Improve data transfer efficiency across accounts

AWS DataSync picked up a monitoring dashboard and shared VPC support. The dashboard shows task status, throughput, duration, and transferred data across an account.

That kind of visibility helps teams spot inefficient or expensive data movement patterns. If transfer jobs are slow, oversized, or repeatedly failing, it becomes much easier to see where time and money are going.

Make reservations and recovery more flexible

AWS gave EC2 reservations and disaster recovery a few practical upgrades. Future-dated Capacity Reservations can now have their start date postponed if plans change.

That matters when demand shifts and you don’t want capacity sitting around unused. More timing flexibility can help avoid stranded commitments and make reservation planning easier to line up with real usage.

Expand model routing and regional options

AWS Bedrock added another model with more routing choices. Grok 4.7 now supports US Geo and Global cross-Region inference options.

That gives teams more room to choose how requests are routed, which can affect cost, performance, and regional processing efficiency. For workloads where placement matters, having more inference options can make optimization easier.