If You Missed It, Here Is What Landed for Artificial Intelligence Builders in September 2026.
Amazon Web Services updated Amazon Bedrock AgentCore with managed agents, efficient runtime, Strands tools, and new OpenAI models for AI builders.
Key Takeaways
- AWS positioned September 2026 Bedrock and AgentCore updates around production readiness: more model choice, lower runtime costs, stronger governance, and better alignment with enterprise data and workflows.
- Amazon Bedrock Managed Agents powered by OpenAI entered public preview, adding durable sessions, human approvals, IAM and CloudTrail controls, and AWS data boundaries for production agent deployment.
- The AgentCore runtime and Strands tools improved efficiency: lower cold starts and pay-as-you-go memory billing, plus a 28% token reduction with Strands harness and a fast open-source Strands Decider 2B for routing and guardrails.
- Bedrock expanded model access across OpenAI, Anthropic, Moonshot, and xAI, including GPT-6 Astra/Sol/Luna, Claude Fable/Opus/Sonnet, Kimi K3, and Grok 4.6/4.7, helping teams balance intelligence, speed, context length, and cost.
- Amazon Bedrock Managed Knowledge Base added automatic daily, weekly, and monthly sync scheduling to keep enterprise agent knowledge current.
Enterprise developers and platform teams reviewing Amazon Bedrock AgentCore updates in September 2026 can find a more practical way to match AI models, runtime costs, and agent governance for production workloads. Amazon Web Services presented the month’s changes as an extension of its broader enterprise AI strategy: builders should have choice at the model layer, the runtime layer, and the tooling layer. The updates touch Amazon Bedrock model access, Amazon Bedrock AgentCore managed infrastructure, and the open source Strands toolkit, with new ways to run agents, reduce token usage, evaluate performance, and keep business information current.
Why the conversation moved past raw model quality
The September changes reflect a shift in how teams evaluate AI systems. Model performance alone no longer decides whether an agent is useful in production. Customers are weighing cost against benefit for specific workloads, especially when agents are assigned tasks that require reasoning over enterprise knowledge, operating across long-horizon work, and acting inside systems where security, accuracy, and governance matter.
In that context, the model becomes only one part of a larger system. Teams also need to consider what context an agent can access, what actions it can take, how its behavior is audited, and how enterprise data is governed and protected. AWS positioned the September updates around those broader needs: more model choice, more efficient agent operations, more measurable behavior, and more direct connections between AI applications and current business information.
Managed agents, faster runtimes, and Strands tools
Amazon Bedrock Managed Agents, powered by OpenAI, entered public preview as a way to build production-ready AI agents while keeping enterprise security controls in place. The offering allows teams to use OpenAI models while keeping data within AWS, reuse existing AWS Identity and Access Management (IAM) permissions, and maintain auditability through AWS CloudTrail. Durable sessions and built-in human approval workflows are intended to reduce development overhead and help mitigate operational risk.
The latest AgentCore runtime, a capability of Amazon Bedrock AgentCore, targets cost and startup performance for serverless agents. It introduces more efficient memory management and lower cold start latency, so teams pay for actual usage rather than peak memory. The runtime supports pay-as-you-go pricing without pre-provisioning capacity. Sessions scale to zero when idle and run in hardware-isolated environments.
Strands received two additions aimed at building agents more efficiently. Strands harness is a new open source agent harness that matches popular harnesses on accuracy while using 28 percent fewer tokens. Teams can start a production-ready agent with a single line of Python or TypeScript, using built-in context management, prompt caching, and memory, then deploy it anywhere.
Strands Decider 2B is a small, open source, 2B-parameter decision model designed to choose between predefined options instead of generating text. It delivers answers in ~115ms locally and is optimized for agentic AI tasks such as tool selection, routing, and guardrails. The code, data, and weights are available on GitHub and Hugging Face.
New model choices for coding, reasoning, and volume
Amazon Bedrock expanded access to OpenAI’s Astra, Sol, and Luna models, which are now generally available. The lineup is intended to cover demanding projects, recurring complex work, and high-volume tasks.
GPT-6 Astra is positioned as the flagship model for ambitious work, supporting complex decisions, document analysis, and software development with up to 1 million input tokens. It is designed to bring stronger reasoning and more precise judgment to business decisions, operate across software and files, and produce professional-quality output aligned with organizational voice, templates, and standards.
GPT-6 Astra Ultrafast adds a premium speed tier for GPT-6 Astra. It is built for workloads where speed matters most, delivering up to 6x faster inference in the API, with up to 300 tokens per second.
Sol is aimed at coding, computer use, and professional workloads that run frequently, bringing near-Astra intelligence to those tasks. Both GPT-6.1 Sol and GPT-6 Sol are available on Amazon Bedrock. GPT 6.1 Luna is designed for high-volume tasks such as extraction, summarization, classification, and routing. Together, these models help teams balance intelligence, latency, and cost across workloads.
Claude options also expanded. Claude Fable 5.1 broadens choices for coding, scientific research, and enterprise workflows. Claude Opus 5.5 is built for agentic coding, knowledge work, and long-running tasks. It uses adaptive thinking to gauge how much reasoning a task needs and includes an effort parameter to set an upper bound on reasoning depth. Claude Sonnet 5.5 is a smarter, more efficient Sonnet for focused coding and knowledge work, delivering 30 percent lower cost per task at 30 percent faster speed than Claude Sonnet 5.
Moonshot AI’s Kimi K3 is now available on Amazon Bedrock. According to Moonshot AI, Kimi K3 is its most capable model and the first open model to reach 2.8 trillion parameters. It includes a 1-million-token context window, native vision, and built-in prompt caching, and is positioned to help teams build smarter apps 2.5x faster at lower cost than previous versions.
xAI’s frontier models Grok 4.6 and Grok 4.7 add another option with a 500K token context window, configurable reasoning effort, and self-verification for more reliable long-running tasks. The combined additions give builders more flexibility when matching models to requirements such as context length, latency, throughput, and cost.
Keeping enterprise agent knowledge current
Amazon Bedrock Managed Knowledge Base gained automatic sync scheduling, allowing teams to configure daily, weekly, and monthly refresh options for all native data-source connectors. The change is meant to help enterprise AI agents retrieve the latest information rather than relying on stale indexed data.
The service also introduced a user-managed setup for SharePoint, OneDrive, and Confluence, reducing reliance on admin-managed service accounts when teams already have the access they need. This can simplify configuration for users who want to connect sources without waiting for centralized administration.
Amazon Bedrock Managed Knowledge Base now supports ServiceNow, Confluence Data Center, Salesforce, and Zendesk as native connectors. These connectors handle data crawling, metadata extraction, and incremental sync automatically, reducing the amount of custom ingestion code teams need to build and maintain for support content, internal documentation, and operational knowledge.
What builders can do next
Teams can explore Amazon Bedrock, deploy agents with the AgentCore CLI, or build a first agent with the Strands Harness SDK. The September 2026 updates are useful for organizations that already use AWS security and governance controls and want to connect model choice, agent runtime efficiency, and enterprise data freshness in one environment.
The recap was published by Prachi Mishra, Senior Product Marketing Manager for Amazon Bedrock AgentCore at Amazon Web Services (AWS), whose role includes bringing production-ready AI agents to developers and enterprises worldwide.
Meta description: Amazon Bedrock AgentCore updates in September 2026 expand OpenAI, Claude, Kimi and Grok model choices while making agents faster and more efficient.
Tags: Amazon Bedrock, AgentCore, AI agents, OpenAI models, Strands
Featured image: An abstract friendly illustration of a cloud, agent icons, model selection dials, and data sync arrows, with no real people, no logos, and no text.

Comments (0)