Amazon Web Services (AWS) has announced a series of significant updates to its Amazon Bedrock and Amazon Bedrock AgentCore services, aimed at empowering AI builders with greater choice, efficiency, and control. The September 2026 enhancements focus on expanding access to leading AI models, improving the performance and cost-effectiveness of AI agents, and providing more robust tools for integrating AI into enterprise workflows.
The Chenab Times has learned that these updates reflect a strategic shift in the industry, moving beyond raw model performance to emphasize the broader system surrounding AI agents. This includes considerations for cost-effectiveness, security, accuracy, and governance, particularly as AI agents are increasingly deployed in production environments to handle complex tasks and reason over enterprise data.
Accelerating Agent Development and Evaluation
A key highlight is the public preview of Amazon Bedrock Managed Agents, which leverage OpenAI models while ensuring enterprise-grade security. This feature allows businesses to utilize OpenAI’s capabilities while keeping their data within the AWS environment, maintaining control through AWS Identity and Access Management (IAM) permissions and auditability via AWS CloudTrail. The introduction of durable sessions and built-in human approval workflows is designed to reduce development overhead and mitigate risks.
Further optimizing agent operations, the updated AgentCore runtime offers more efficient memory management and reduced cold start latency for serverless agents. This translates to faster agent execution and lower costs, with a pay-as-you-go pricing model that eliminates the need for pre-provisioning capacity. Sessions are designed to scale to zero when idle and operate in hardware-isolated environments, enhancing both performance and cost efficiency.
For developers seeking to minimize token consumption without compromising accuracy, Strands harness, a new open-source agent harness, has been introduced. This tool reportedly matches popular harnesses on accuracy while utilizing 28 percent fewer tokens. It enables the rapid deployment of production-ready agents with built-in context management, prompt caching, and memory, accessible via a single line of Python or TypeScript code and deployable across various environments.
Complementing these advancements, Strands Decider 2B, an open-source, 2B-parameter decision model, offers localized, low-latency responses for agentic AI tasks such as tool selection and routing. Optimized for speed and accuracy, its code, data, and weights are available on GitHub and Hugging Face.
Expanding Model Diversity and Performance
AWS has broadened the selection of models available on Amazon Bedrock to cater to diverse workload requirements. OpenAI’s Astra, Sol, and Luna models are now generally available, offering options for demanding projects, recurring complex tasks, and high-volume operations.
GPT-6 Astra, positioned as a flagship model for ambitious projects, provides enhanced reasoning for complex decisions, document analysis, and software development, supporting up to one million input tokens. An accelerated version, GPT-6 Astra Ultrafast, offers up to six times faster inference for latency-sensitive workloads, reaching speeds of up to 300 tokens per second.
GPT-6.1 Sol and GPT-6 Sol are available for coding, professional tasks, and frequent computer use, bringing near-Astra intelligence. GPT 6.1 Luna is specifically designed for high-volume tasks like extraction and summarization, enabling a balance between intelligence, latency, and cost.
For coding, scientific research, and enterprise workflows, updated Claude models are now accessible. Claude Fable 5.1 enhances options for coding and research, while Claude Opus 5.5 is geared towards agentic coding and long-running tasks, featuring adaptive thinking to adjust reasoning depth. Claude Sonnet 5.5 provides a more intelligent and efficient option for focused coding and knowledge work, offering reduced costs and faster speeds compared to its predecessor.
The portfolio also includes Moonshot AI’s Kimi K3, described as its most capable model with 2.8 trillion parameters, a one-million-token context window, and native vision capabilities. Additionally, xAI’s Grok 4.6 and Grok 4.7 offer a 500K token context window and support configurable reasoning and self-verification for improved reliability in long-running tasks.
Enhanced Control for Enterprise AI Agents
Updates to Amazon Bedrock Managed Knowledge Base introduce automatic sync scheduling, allowing daily, weekly, or monthly refresh options for data sources to ensure AI agents access the most current information. User-managed setup for platforms like SharePoint, OneDrive, and Confluence simplifies integration by reducing reliance on admin-managed service accounts.
Furthermore, Amazon Bedrock Managed Knowledge Base now natively supports ServiceNow, Confluence Data Center, Salesforce, and Zendesk. These new connectors automate data crawling, metadata extraction, and incremental synchronization, significantly reducing the need for custom ingestion pipelines and facilitating the integration of support content, internal documentation, and operational knowledge.
Developers can explore these new capabilities through Amazon Bedrock, deploy agents using the AgentCore CLI, or build agents with the Strands Harness SDK. AWS encourages interested teams to connect with them to discuss how Amazon Bedrock can support their AI initiatives.
The Chenab Times News Desk