ReleaseBytes
anthropic Claude Platform Release Notes ·

Anthropic Claude Platform Updates: Opus 5, Agent Enhancements, & Deprecations

aigaengineer
feature breaking deprecation

Anthropic has released extensive updates to the Claude Platform, API, SDKs, and Console throughout August 2026. Key changes include the launch of Claude Opus 5 with a 1M token context window and new effort controls, alongside significant enhancements for Claude Managed Agents such as session budgets, advisor models, and inference geo control. The Compliance API has been expanded with new session transcript capabilities for Enterprise organizations, and several features like Claude Opus 4.1 and the legacy Workbench are being retired. These updates provide developers with more powerful models, greater control over agent behavior, and improved compliance tools.

  • →Claude Opus 5 Model Launched with Enhanced Capabilities
  • →Significant Enhancements for Claude Managed Agents
  • →Expanded Compliance API and Security Controls for Enterprise
  • →Multiple Models and Experimental Features Retired
  • →Claude Sonnet 5 Introductory Pricing Standardized
Deprecations (1) ›
  • Multiple Models and Experimental Features Retired

    Claude Opus 4.1 has been retired, with all requests now returning errors. Fast mode for Claude Opus 4.7 is also removed, requiring migration to newer Opus versions for similar functionality. The legacy Workbench in the Claude Console and experimental prompt tools APIs are being sunset, with access ending on August 17, 2026.

Features (3) ›
  • Claude Opus 5 Model Launched with Enhanced Capabilities

    Claude Opus 5, offering a 1M token context window and default thinking, is now available across various platforms. Disabling its thinking capability is restricted to effort levels high or below, a breaking change from previous Opus versions. The Dreams research preview also now supports Claude Opus 5.

  • Significant Enhancements for Claude Managed Agents

    Managed Agents now support session budgets for spend control and can consult advisor models for strategic guidance. Users can specify inference geographies, load skills directly from GitHub repositories, and set agent-level effort configurations. New webhooks and initial event seeding further expand control over agent lifecycle and interaction.

  • Expanded Compliance API and Security Controls for Enterprise

    The Compliance API now provides beta access for Enterprise organizations to transcripts of Cowork and Claude Code sessions, both local and remote. Additionally, inference hooks are in beta, allowing organizations to route governed prompts through their AI security server before inference, and a new Admin API facilitates member management for Enterprise organizations.

Enhancements (1) ›
  • API Improvements for Conversation Flow and Tooling

    Mid-conversation tool and system message changes are now available on select Claude models, allowing dynamic adjustments during a conversation. The fallbacks parameter now includes a 'default' mode with recommended models by refusal category, and the agent update API's version field is now optional for unconditional updates.

Notes (1) ›
  • Claude Sonnet 5 Introductory Pricing Standardized

    The introductory pricing for Claude Sonnet 5 ($2 / $10 per MTok) has been made the standard price. The previously scheduled price increase will no longer occur, providing cost consistency for users.

Read the original announcement →

https://platform.claude.com/docs/en/release-notes/overview

Related releases