A recap of the most recent Amazon Bedrock, Amazon Bedrock AgentCore, and Strands updates from September 2026
At AWS, we imagine the quickest path to enterprise AI is giving builders actual selection at each layer: mannequin, runtime, and tooling. That perception shapes how we make investments throughout Amazon Bedrock, AgentCore, and Strands. Amazon Bedrock continues increasing entry to main frontier fashions so you’ll be able to construct and scale AI functions securely. Amazon Bedrock AgentCore offers the managed infrastructure to construct, join, and deploy enterprise brokers utilizing any framework or mannequin. Strands offers you an open supply toolkit to construct brokers and run wherever.
As AI fashions turn out to be extra succesful and mannequin selection expands, the business conversations are shifting. Mannequin efficiency is now not the one query. Clients now weigh value towards profit for his or her particular use case, and people choices matter when delegating work to brokers in manufacturing. AI brokers are already transferring into workflows that require them to motive over enterprise information, function throughout long-horizon duties, and act in programs the place safety, accuracy, and governance matter. The main target is now shifting past the mannequin itself to the broader system round it, together with the context brokers can entry, the actions they’ll take, and the way enterprise information is ruled and guarded.
In September, updates throughout Amazon Bedrock, AgentCore, and Strands strengthened these foundations. New capabilities expanded mannequin selection, made agent operations extra environment friendly and measurable, and created extra direct methods to attach AI functions with present enterprise info.
Run sooner brokers and consider what issues
Construct brokers optimized for OpenAI fashions. Amazon Bedrock Managed Brokers, powered by OpenAI, is now obtainable in public preview, serving to you construct production-ready AI brokers with out sacrificing enterprise safety. Now you can use OpenAI fashions whereas protecting information inside AWS, reuse current AWS Id and Entry Administration (IAM) permissions, and keep full auditability via AWS CloudTrail. Sturdy periods and built-in human approval workflows assist scale back growth overhead and mitigate threat.
Brokers now begin sooner with much less infrastructure overhead. The newest AgentCore runtime, a functionality of Amazon Bedrock AgentCore, has extra environment friendly reminiscence administration and decrease chilly begin latency for serverless brokers, so that you pay for precise utilization fairly than the height reminiscence. You may run brokers sooner at decrease value with pay-as-you-go pricing and no have to pre-provision capability. Classes scale to zero when idle and run in hardware-isolated environments.
Decrease your token value with frontier efficiency. Strands harness is a brand new open supply agent harness that matches fashionable harnesses on accuracy whereas utilizing 28 % fewer tokens. You may spin up a production-ready agent in a single line of Python or TypeScript, with built-in context administration, immediate caching, and reminiscence, then deploy it wherever.
Let your fashions make massive choices safely, regionally. Strands Decider 2B is a small, open supply, 2B-parameter resolution mannequin that picks between predefined choices (as an alternative of producing textual content), delivering solutions in ~115ms regionally. It’s optimized for agentic AI duties like software choice, routing, and guardrails, with all code, information, and weights obtainable on GitHub and Hugging Face.
Discover the proper mannequin for every workload
Select from a broader set of OpenAI fashions in your on a regular basis work. OpenAI’s Astra, Sol, and Luna fashions are actually usually obtainable on Amazon Bedrock, with choices for demanding initiatives, recurring complicated work, and high-volume duties.
GPT-6 Astra is the flagship mannequin in your most bold work, bringing deeper reasoning to complicated choices, doc evaluation, and software program growth with as much as 1 million enter tokens. It brings stronger reasoning and extra exact judgment to complicated enterprise choices, operates throughout software program and information, and delivers professional-quality output aligned with organizational voice, templates, and requirements. GPT-6 Astra Ultrafast provides a premium pace tier for GPT-6 Astra, constructed for workloads the place pace issues most. It delivers as much as 6x sooner inference within the API, with as much as 300 tokens per second.
Sol brings near-Astra intelligence to coding, pc use, {and professional} workloads that run often. Each GPT-6.1 Sol and GPT-6 Sol can be found on Amazon Bedrock. GPT 6.1 Luna is designed for high-volume duties corresponding to extraction, summarization, classification, and routing. Collectively, they assist you steadiness intelligence, latency, and value throughout workloads.
Tackle long-running coding and analysis with Claude. Mannequin selections for Claude now embody the most recent variations to advance your coding and scientific work. Claude Fable 5.1 expands choices for coding, scientific analysis, and enterprise workflows. Claude Opus 5.5 is constructed for agentic coding, information work, and long-running duties. It makes use of adaptive pondering to gauge how a lot reasoning a job wants and an effort parameter to set an higher certain on reasoning depth. Claude Sonnet 5.5 is a wiser, extra environment friendly Sonnet suited to targeted coding and information work, delivering 30 % decrease value per job at 30 % sooner pace than Claude Sonnet 5.
Purpose via complexity with a broader mannequin portfolio. Moonshot AI’s Kimi K3 is now obtainable on Amazon Bedrock. In keeping with Moonshot AI, Kimi K3 is its most succesful mannequin and the primary open mannequin to achieve 2.8 trillion parameters. With a 1-million-token context window, native imaginative and prescient, and built-in immediate caching, it helps you construct smarter apps 2.5x sooner at decrease value than earlier variations.
xAI’s frontier fashions Grok 4.6 and Grok 4.7 provide a 500K token context window and help configurable reasoning effort and self-verification for extra dependable long-running duties.
Collectively, these additions provide you with extra flexibility when matching fashions to your necessities corresponding to context size, latency, throughput, and value.
Get extra management over your enterprise AI brokers
Hold agent information present. With new computerized sync scheduling in Amazon Bedrock Managed Data Base, now you can configure each day, weekly, and month-to-month refresh choices for all native data-source connectors. In consequence, your enterprise AI brokers retrieve the most recent info. The user-managed setup for SharePoint, OneDrive, and Confluence reduces reliance on admin-managed service accounts when you have already got the entry you want.
Join extra enterprise sources with out sustaining customized ingestion pipelines. Amazon Bedrock Managed Data Base now helps ServiceNow, Confluence Information Middle, Salesforce, and Zendesk as native connectors to deal with information crawling, metadata extraction, and incremental sync mechanically. These connectors scale back the quantity of customized ingestion code that you must construct and keep help content material, inside documentation, and operational information.
Get began
Discover Amazon Bedrock, deploy brokers with the AgentCore CLI, or construct your first agent with the Strands Harness SDK.
Taken with studying how Amazon Bedrock can help your crew? Join with us to start out the dialog.
Concerning the creator
