ReleaseBytes
anthropic Claude Platform Release Notes ·

Anthropic Updates Claude Platform with Opus 5, Managed Agents Features, and Enterprise Tools

aipreviewengineer
feature breaking deprecation announcement

Anthropic has launched Claude Opus 5, featuring a 1M token context window and significantly enhanced capabilities for developers. The platform also gained new features for Claude Managed Agents, including session budgets, advisors, and data residency controls, alongside significant updates to enterprise compliance and security tools like Inference Hooks and an expanded Compliance API. These updates enhance model performance and control for developers, provide robust management and auditing capabilities for enterprise users, and expand the flexibility of agent-based applications. Opus 5 also introduces a breaking change regarding disabling its "thinking" feature at higher effort levels, and the company adjusted Claude Sonnet 5 pricing while deprecating older models and the legacy Workbench.

  • →Claude Opus 5 Launched with 1M Token Context and Enhanced Thinking
  • →New Managed Agents Capabilities: Budgets, Advisors, Data Residency & GitHub Skills
  • →Expanded Enterprise Compliance and Security Features
  • →Sonnet 5 Pricing Standardized; Legacy Models & Workbench Retired
  • →Mid-Conversation Tools, System Messages, and API Fallback Modes
Deprecations (1) ›
  • Sonnet 5 Pricing Standardized; Legacy Models & Workbench Retired

    The introductory pricing for Claude Sonnet 5 is now the standard rate, with the previously scheduled increase cancelled. Claude Opus 4.1 has been retired, with requests now returning errors. Fast mode for Claude Opus 4.7 has also been removed. The legacy Workbench and experimental prompt tools APIs are being sunset, with access ending August 17, 2026.

Features (3) ›
  • Claude Opus 5 Launched with 1M Token Context and Enhanced Thinking

    Claude Opus 5 is now available, featuring a 1M token context window, 128k max output tokens, and thinking enabled by default. This release also introduces a breaking change where disabling thinking at `xhigh` or `max` effort levels will return an error. Dreams (research preview) now supports Claude Opus 5.

  • New Managed Agents Capabilities: Budgets, Advisors, Data Residency & GitHub Skills

    Claude Managed Agents can now be configured with session budgets to cap spend and utilize an advisor model for strategic guidance. Users can control inference data residency and load skills directly from GitHub repositories. Additional updates include setting agent effort levels, new webhooks for environment and memory store lifecycle events, and the ability to seed sessions with initial events.

  • Expanded Enterprise Compliance and Security Features

    Inference hooks are now in beta for Claude Enterprise organizations, allowing prompts to be held for AI security server verdicts. The Compliance API has been expanded (in beta) to return transcripts of Cowork and Claude Code sessions, both local and remote, for auditing purposes. Additionally, an Admin API is in beta for managing users within Claude Enterprise organizations.

Enhancements (1) ›
  • Mid-Conversation Tools, System Messages, and API Fallback Modes

    Mid-conversation tool changes are in beta, allowing tools to be added or removed between turns. Mid-conversation system messages are now generally available on specific models without a beta header. The `fallbacks` parameter now supports a "default" mode with server-side fallback in beta.

Read the original announcement →

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

Related releases