AWS Adds OpenAI, Claude, Kimi K3 and Grok Models to Bedrock in September

Amazon Web Services used its September 2026 recap to detail a broad set of updates across Amazon Bedrock, Amazon Bedrock AgentCore and the open source Strands toolkit. They add new models from OpenAI, Anthropic, Moonshot AI and xAI, along with cheaper, faster agent runtimes and more connectors for enterprise data.
- Amazon Bedrock Managed Agents, powered by OpenAI, is in public preview, with data kept within AWS and IAM and CloudTrail support.
- OpenAI's Astra, Sol and Luna models are generally available on Bedrock, including GPT-6 Astra Ultrafast with up to 300 tokens per second.
- Claude Fable 5.1, Claude Opus 5.5 and Claude Sonnet 5.5, plus Kimi K3, Grok 4.6 and Grok 4.7, join the Bedrock model lineup.
- Strands harness uses 28 percent fewer tokens than popular harnesses while matching accuracy, according to AWS.
- Managed Knowledge Base adds scheduled syncs and connectors for ServiceNow, Confluence Data Center, Salesforce and Zendesk.
Agents: OpenAI support, faster runtime and a leaner harness
According to AWS, Amazon Bedrock Managed Agents, powered by OpenAI, is now in public preview. It lets teams use OpenAI models while keeping data within AWS, reuse existing AWS Identity and Access Management (IAM) permissions and keep full auditability through AWS CloudTrail. Durable sessions and built-in human approval workflows are meant to reduce development overhead and risk.
The latest AgentCore runtime has more efficient memory management and lower cold start latency for serverless agents. AWS says customers pay for actual usage rather than peak memory, with pay-as-you-go pricing and no need to pre-provision capacity. Sessions scale to zero when idle and run in hardware-isolated environments.
On the open source side, AWS describes the new Strands harness as matching popular harnesses on accuracy while using 28 percent fewer tokens. A production-ready agent can be started in a single line of Python or TypeScript, with built-in context management, prompt caching and memory. Strands Decider 2B is a small open source decision model with 2B parameters. It picks between predefined options instead of generating text and returns answers in about 115ms locally, according to AWS. It targets tasks such as tool selection, routing and guardrails, and its code, data and weights are on GitHub and Hugging Face.
A wider model lineup on Bedrock
OpenAI's Astra, Sol and Luna models are now generally available on Amazon Bedrock. GPT-6 Astra is the flagship, aimed at complex decisions, document analysis and software development, with up to 1 million input tokens. GPT-6 Astra Ultrafast is a premium speed tier that delivers up to 6x faster inference in the API, with up to 300 tokens per second. Both GPT-6.1 Sol and GPT-6 Sol are available, positioned for coding, computer use and professional workloads that run frequently. GPT 6.1 Luna is designed for high-volume tasks such as extraction, summarization, classification and routing.
For Anthropic's Claude, AWS lists Claude Fable 5.1 for coding, scientific research and enterprise workflows, and Claude Opus 5.5 for agentic coding, knowledge work and long-running tasks. Opus 5.5 uses adaptive thinking to gauge how much reasoning a task needs, plus an effort parameter that sets an upper bound on reasoning depth. Claude Sonnet 5.5 delivers 30 percent lower cost per task at 30 percent faster speed than Claude Sonnet 5, according to AWS.
Moonshot AI's Kimi K3 is also available. According to Moonshot AI, it is the company's most capable model and the first open model to reach 2.8 trillion parameters. It offers a 1-million-token context window, native vision and built-in prompt caching. xAI's Grok 4.6 and Grok 4.7 provide a 500K token context window and support configurable reasoning effort and self-verification.
Fresher knowledge and more connectors
Amazon Bedrock Managed Knowledge Base now has automatic sync scheduling, with daily, weekly and monthly refresh options for all native data-source connectors, so agents retrieve the latest information. A user-managed setup for SharePoint, OneDrive and Confluence reduces reliance on admin-managed service accounts.
The service also adds native connectors for ServiceNow, Confluence Data Center, Salesforce and Zendesk. They handle data crawling, metadata extraction and incremental sync automatically, which cuts the custom ingestion code teams need to build and maintain.
What it means for builders
AWS frames the updates around a shift in enterprise AI: model performance is no longer the only question, and customers now weigh cost against benefit for each use case. The company says attention is moving to the system around the model, including the context agents can access, the actions they can take and how enterprise data is governed.
For teams building on AWS, the releases offer more choice when matching models to context length, latency, throughput and cost, along with lower-overhead runtimes and managed data connections. Starting points named by AWS are Amazon Bedrock, the AgentCore CLI and the Strands Harness SDK.


