Frontier AI
Cost Explorer, Budgets and Cost Management Dashboards can group and filter Bedrock costs by model, model provider, inference type and feature.
Why it matters. You can now set a budget per model or provider without custom tagging, at no extra charge, outside GovCloud and the China Regions.
Source: AWS What's New
Enterprise administrators can pick any available Copilot agent model as the default for new conversations, and stop MCP servers starting automatically.
Why it matters. The update needs JetBrains IDE 2025.2 or later, so teams still on 2025.1 must upgrade their IDE to get it.
Source: GitHub changelog
Copilot Pro, Pro+, Max, Business and Enterprise users can now choose Claude Haiku 5.5.
Why it matters. Teams that want a faster, cheaper Claude model for routine Copilot work can now select it directly.
Source: GitHub Copilot weekly releases
The deep-research-pro-preview-12-2025 agent is deprecated, with deep-research-preview-04-2026 and deep-research-max-preview-04-2026 as replacements.
Why it matters. Change the agent parameter in your interactions.create calls before October 23 or those requests stop working.
Source: Gemini API release notes
The Responses API on Bedrock now accepts the reasoning.summary parameter for OpenAI GPT models, in Region and cross-Region inference.
Why it matters. You can debug agent answers or show users how the model reached them without leaving Bedrock.
Source: AWS What's New
Ultrafast mode for GPT-6.1 Sol, which speeds up inference, is now on Amazon Bedrock through the console and supported Bedrock APIs.
Why it matters. Teams that buy OpenAI models through AWS can now test the faster tier for coding assistants and interactive agents.
Source: AWS What's New
In beta for Claude Enterprise, the chat endpoints return unified experience chats with your existing key, and Claude Docs documents download as Word files.
Why it matters. Compliance and eDiscovery exports now cover these chats, so check that your retention tooling picks them up.
Source: Claude Platform release notes
Both models are now in the Kiro IDE and CLI in the AWS GovCloud (US) Regions, with 1M context windows.
Why it matters. Opus 5.5 uses a 2.0x credit multiplier, down from 2.2x for Opus 5, so update and restart Kiro to switch.
Source: AWS What's New