ReleaseBytes
anthropic Claude Platform Release Notes ·

Anthropic Launches Claude Opus 5 with 1M Token Context and Enhanced Controls

aigapreviewengineer
feature breaking deprecation

Anthropic has launched Claude Opus 5, a significant upgrade to its large language model family, featuring a 1M token context window, 128k max output tokens, and "thinking" enabled by default. This new model maintains the same pricing as Opus 4.8 and is available across the Claude API and various cloud platforms. Additionally, the platform introduces beta features like mid-conversation tool changes and a "default" server-side fallback mode, while deprecating fast mode for Claude Opus 4.7. These updates offer developers more powerful capabilities, refined control, and improved error handling for their AI applications.

  • →Claude Opus 5 model launched with expanded context
  • →Breaking change for Claude Opus 5 'thinking' parameter
  • →Mid-conversation tool changes introduced in beta
  • →Server-side fallback with 'default' mode in beta
  • →Fast mode removed for Claude Opus 4.7
Breaking changes (1) ›
  • Breaking change for Claude Opus 5 'thinking' parameter

    On Claude Opus 5, disabling the "thinking" parameter is now restricted to effort `high` or below. Attempts to use `thinking: {"type": "disabled"}` with effort `xhigh` or `max` will result in a 400 error, which is a breaking change from Claude Opus 4.8.

Deprecations (1) ›
  • Fast mode removed for Claude Opus 4.7

    Fast mode has been removed for Claude Opus 4.7, meaning requests with `speed: "fast"` to `claude-opus-4-7` will now return an error. Users are advised to migrate to Claude Opus 5 or Claude Opus 4.8 to continue utilizing fast mode.

Features (3) ›
  • Claude Opus 5 model launched with expanded context

    Anthropic has launched Claude Opus 5 (`claude-opus-5`), an upgrade to Opus 4.8, supporting a 1M token context window and 128k max output tokens. It features "thinking" on by default at the same pricing as its predecessor and is available on the Claude API and major cloud platforms.

  • Mid-conversation tool changes introduced in beta

    Beta support for mid-conversation tool changes is now available on Claude Fable 5, Claude Mythos 5, Claude Opus 4.8, and Claude Opus 5. This allows tools to be added or removed between conversation turns while preserving the prompt cache, and requires a specific beta header.

  • Server-side fallback with 'default' mode in beta

    The `fallbacks` parameter now supports a "default" mode, applying Anthropic's recommended fallback models by refusal category. This server-side fallback is in beta and requires the `server-side-fallback-2026-07-01` beta header for activation.

Enhancements (1) ›
  • Enhanced 'Effort' control for Claude Opus 5

    The "Effort" parameter is now the primary control for steering Claude Opus 5, supporting the full range of `low`, `medium`, `high`, `xhigh`, and `max`. The `max` setting is specifically designed for capability-critical workloads.

Read the original announcement →

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

Related releases