AWS AI Updates: October 10, 2026

1. TwelveLabs Pegasus 1.5 Video-to-Text Model Arrives on Amazon Bedrock

AWS. Amazon Bedrock now offers TwelveLabs Pegasus 1.5, a video-to-text model that produces structured, time-coded metadata from entire videos. Users define the moments that matter, such as editorial segments, speaker changes, sports plays, or brand appearances, and the model returns them with timestamps after analyzing visuals, audio, and on-screen text in a single pass with no pre-indexing pipeline. The model is accessible through the US and Global cross-Region inference profiles. Source

2. Bedrock Adds Reasoning Summaries for OpenAI Models via the Responses API

AWS. Amazon Bedrock now supports the reasoning.summary parameter for OpenAI models through the Responses API, returning a human-readable summary of the model’s reasoning in the summary array of the reasoning output item alongside the answer. AWS positions the feature for evaluating responses, debugging applications, and showing users how a model approached coding, analysis, or multi-step tasks. It works for all OpenAI models on Bedrock across in-Region, geographic cross-Region, and global cross-Region inference. Source

3. Postman Details How Agent Mode Runs on Bedrock for 40 Million Developers

AWS. A new AWS Machine Learning Blog post describes how Postman built Agent Mode on Anthropic Claude models in Amazon Bedrock, using a vector index of tool embeddings to narrow more than 170 tools to about 15 per request after tool-selection errors rose beyond roughly 40 visible tools. The architecture combines a RAG knowledge base built from Postman’s Learning Center, a single schema-based ClickHouse query tool for API Catalog reads, and user approval before any state-changing action. On the Bedrock side, Postman uses geographic and global cross-Region inference profiles, zero data retention where supported, layered prompt caching with one-hour and five-minute TTL checkpoints, and optional Guardrails for PII redaction. Source

4. SageMaker Unified Studio Lets Admins Define Projects With Custom CloudFormation Blueprints

AWS. Amazon SageMaker Unified Studio now supports custom Tooling blueprints, allowing domain administrators to register their own AWS CloudFormation templates as the foundation for every project. Templates can include any CloudFormation resource, such as Lake Formation grants, Athena workgroups, or VPC security groups, and can enforce company IAM naming standards or custom permissions boundaries. The service validates templates at registration and after each deployment, and auto-populates reserved parameters such as project ID and domain ID so one template works across accounts and Regions. Source

5. Amazon Connect Flags Evaluation Questions That AI Cannot Score Reliably

AWS. Amazon Connect Customer now offers a one-click automated check that compares performance evaluation forms against best practices and lists questions lacking enough context for AI to score them accurately. The check suggests more specific wording, for example prompting a manager to define exactly what a required sales disclosure must contain. The feature is aimed at scoring both human and AI agents consistently and is available in eight Regions across North America, Europe, and Asia Pacific. Source

6. Amazon Quick Adds Brand Templates for AI-Generated Presentations and Documents

AWS. Amazon Quick now lets brand managers upload PowerPoint or Word templates, or pull them from connected stores such as SharePoint, and extracts colors, fonts, logos, layouts, and brand rules for review before publishing them organization-wide. When generating new assets, Quick selects the best-fit template based on manager guidance, follows its rules, and explains and suggests alternatives when a request would break them. Administrators control who can create and publish templates through custom permissions, and the feature is generally available in the Quick desktop app. Source