# AWS Expands AI Agent Capabilities and Adds New Frontier Models to Bedrock Platform
Amazon Web Services has unveiled a sweeping set of updates across its cloud ecosystem, marking a significant week for developers, enterprises, and AI practitioners alike. From a major expansion of agent-building tools to the addition of cutting-edge language models, these announcements reflect AWS’s growing commitment to making advanced AI more accessible, secure, and production-ready within its own infrastructure.
## Amazon Bedrock Managed Agents Now Powered by OpenAI
One of the headline announcements is the public preview of Amazon Bedrock Managed Agents powered by OpenAI. This new offering is built on a customized version of OpenAI’s Agents API, specifically engineered to operate as an AWS-native service. What does this mean in practical terms? Organizations can now build sophisticated AI agents that leverage OpenAI’s model capabilities while running entirely inside AWS — complete with the familiar identities, permissions, and governance frameworks they already rely on for compliance and security.
Developers have the flexibility to choose their execution environment. Those who prefer to keep things on their own terms can opt for self-hosted compute, allowing them to use existing development machines, containers, or custom compute environments. Alternatively, Amazon Bedrock AgentCore Runtime offers a managed approach with configurable runtime sessions and storage, all within the AWS account. This dual-path design ensures that teams can match their deployment strategy to their operational needs without sacrificing performance or control.
## Four New Frontier Models Broaden the Bedrock Catalog
Alongside the agent infrastructure, AWS has added four new models to the Amazon Bedrock portfolio, giving developers more options for specific workloads.
**OpenAI GPT-6.1 Sol** represents a significant upgrade over its predecessor, GPT-6 Sol. It delivers top-tier performance across agentic coding, computer use, and professional work scenarios. Notably, it approaches the capabilities of GPT-6 Astra on demanding benchmarks while costing roughly one-fifth as much — a compelling proposition for teams looking to scale agentic workloads without a proportional increase in compute costs.
**OpenAI GPT-6 Astra UltraFast** is a premium speed tier designed for latency-sensitive applications. According to OpenAI, UltraFast delivers up to six times faster inference compared to standard tiers, achieving throughput of up to 300 tokens per second. The underlying Amazon Bedrock inference engine ensures that this raw speed is accompanied by the security, reliability, and production-grade durability that enterprise workloads demand.
**Anthropic Claude Sonnet 5.5** is a refined iteration of the Sonnet family. It offers improved coding intelligence and is particularly well-suited for completing well-scoped tasks as part of a larger development strategy. Teams already building on Sonnet will find it to be a natural and efficient upgrade path, with stronger capabilities for building features, fixing bugs, and verifying output against defined requirements — all within a single session.
**SpaceXAI Grok 4.7** builds on the foundation of Grok 4.6 with enhancements in mixed-document handling, repo-scale coding with planning and error recovery, and improved browser-use agents capable of handling form fills and portal navigation. These improvements make Grok 4.7 a strong contender for scenarios involving complex, multi-step web interactions and large-scale codebases.
## Notable Service Updates and Launches
Beyond the AI model and agent announcements, several other updates caught attention during the week.
The **AWS Well-Architected Agent** (now in preview) introduces an AI-powered service that scans your AWS environment and delivers targeted, contextual recommendations to improve cost efficiency, security posture, application performance, and overall resilience. Rather than relying on manual reviews, this agent analyzes your infrastructure and aligns its suggestions with your unique business objectives.
For data engineers and database professionals, **Amazon Aurora PostgreSQL** now supports direct querying of Apache Iceberg and Parquet data. This means operational data can be queried alongside data lake assets using existing PostgreSQL tools and applications — eliminating the need for complex ETL pipelines or data duplication, streamlining the path from raw storage to actionable insights.
**Amazon S3 Tables** have also expanded their capabilities, now supporting all Apache Iceberg V3 data types. This includes native support for geometry, geography, unknown, and nanosecond timestamp data types, as well as column default values. The practical impact is that teams can now store geospatial coordinates and ultra-precise event timestamps directly in their tables, rather than resorting to string or integer encodings that complicate querying and analysis.
## Broader AWS Ecosystem Developments
Several other noteworthy developments round out the week’s announcements.
**Kiro workflows** introduce a way to execute complex, multi-step tasks using multiple agents with reduced oversight. This feature is being integrated into Kiro itself, covering cloud configuration, cloud sessions, and the broader workflow experience — making it easier to automate end-to-end processes across AWS resources.
**Strands Decider** represents a new class of model known as a decision model, or “system one” model. Designed for fast experimentation, local development, and rapid innovation, Strands Decider 2B is small, open source, and optimized for iterative exploration — complementing larger, more resource-intensive models in the development pipeline.
AWS also expanded its **Forward Deployed Engineering (FDE)** program to partners. Originally backed by a $1 billion investment, FDE now includes three new Partner FDE pathways and credentials that formally recognize the applied proficiency required to deliver production-grade agentic AI solutions. This structured approach helps partners validate their expertise and provides enterprises with a clearer way to identify qualified implementation partners.
—
## Frequently Asked Questions (FAQ)
**Q1: What is Amazon Bedrock Managed Agents powered by OpenAI?**
A: It is a new AWS service that allows developers to build AI agents using OpenAI’s Agents API, fully integrated with AWS-native security, identity, and governance controls. Agents can be executed in either self-hosted compute environments or via the managed Bedrock AgentCore Runtime.
**Q2: How does GPT-6.1 Sol compare to GPT-6 Astra in terms of cost?**
A: GPT-6.1 Sol delivers performance that approaches GPT-6 Astra on demanding evaluations but at roughly one-fifth of the cost, making it a more economical option for large-scale agent deployments.
**Q3: What makes GPT-6 Astra UltraFast different from the standard GPT-6 Astra model?**
A: UltraFast is a premium speed tier that delivers up to six times faster inference, with throughput reaching up to 300 tokens per second, making it ideal for latency-sensitive production workloads.
**Q4: Who should upgrade to Claude Sonnet 5.5?**
A: Teams already building on Claude Sonnet will benefit most from this upgrade. It offers stronger coding capabilities and improved task completion within larger development strategies, all within the same session context.
**Q5: Can I query Apache Iceberg and Parquet data directly from Aurora PostgreSQL?**
A: Yes. Amazon Aurora PostgreSQL now supports direct querying of both Apache Iceberg and Parquet data formats, allowing you to combine operational and data lake data without ETL pipelines or data duplication.
**Q6: What are Kiro workflows designed for?**
A: Kiro workflows enable developers to carry out complex, multi-agent tasks from start to finish with minimal supervision, covering areas like cloud configuration and session management.
**Q7: Is Strands Decider open source?**
A: Yes. Strands Decider 2B is a small, open-source decision model optimized for fast experimentation and local development.
**Q8: What are the FDE Partner pathways?**
A: The Forward Deployed Engineering organization now offers three new Partner FDE pathways and credentials that validate a partner’s applied proficiency in delivering production agentic AI solutions.
—
## Conclusion
This week’s AWS announcements signal a decisive push toward making agentic AI more practical, affordable, and deeply integrated within the cloud ecosystem. The combination of OpenAI-powered managed agents, a diversified roster of new frontier models, and expanded data querying capabilities across AWS services gives developers and enterprises a powerful and flexible toolkit. Whether the goal is to reduce inference costs, accelerate production deployments, or automate complex workflows, AWS has laid out a clear path forward. As these services move from preview to general availability, the impact on how teams build, deploy, and govern AI applications in the cloud is expected to be substantial.
Thank you for reading
