ICYMI: What landed for AI builders in September 2026

Model Choice Expansion and Performance Enhancements

Amazon Bedrock has expanded its model portfolio with the general availability of OpenAI’s Astra, Sol, and Luna models. These models cater to different workloads, from high-complexity tasks with Astra to high-volume operations with Luna. The addition of Moonshot AI’s Kimi K3 and xAI’s Grok 4.6 and 4.7 further diversifies the available options, offering capabilities such as 2.8 trillion parameters and 500K token context windows. These updates provide builders with more flexibility in selecting models that align with specific requirements like context length, latency, and cost.

Agent Efficiency and Evaluation

Amazon Bedrock AgentCore has introduced a new runtime with improved memory management and reduced cold start latency, enabling faster agent execution and lower costs. The Strands Harness, an open-source agent framework, reduces token usage by 28% while maintaining accuracy, making it more efficient for production deployment. Additionally, Strands Decider 2B, a lightweight decision model, offers fast, local inference for tasks like tool selection and guardrails, enhancing agent performance without relying on external cloud resources.

Enterprise Integration and Knowledge Syncing

Amazon Bedrock Managed Knowledge Base now supports native connectors for enterprise systems like ServiceNow, Confluence Data Center, Salesforce, and Zendesk. These integrations automate data crawling, metadata extraction, and incremental syncing, reducing the need for custom ingestion pipelines. Automatic sync scheduling allows users to configure daily, weekly, or monthly refreshes, ensuring that AI agents have access to the most up-to-date enterprise data without manual intervention.

Security and Governance

Amazon Bedrock Managed Agents, powered by OpenAI, are now available in public preview, allowing builders to use OpenAI models within AWS while maintaining enterprise security. These agents leverage existing AWS IAM permissions and provide full auditability through AWS CloudTrail. This integration ensures that data remains within AWS, reducing exposure and enhancing compliance for organizations handling sensitive information.

Practical Guidance for Builders

When evaluating these updates, builders should consider the tradeoffs between model performance, cost, and integration complexity. For instance, while Kimi K3 offers advanced capabilities, its large parameter count may increase inference costs. Similarly, while Strands Decider 2B provides fast local inference, it may not be suitable for tasks requiring extensive reasoning. Builders should also assess the compatibility of new models with existing workflows and infrastructure, particularly when integrating with enterprise systems like Salesforce or Confluence. Finally, the use of automated knowledge syncing can reduce maintenance overhead but requires careful configuration to avoid data duplication or inconsistency.

Source: AWS Machine Learning. Analysis based on publicly available information; no hands-on testing is claimed.