Claude Platform Updates: Opus 4.8, Mid-conversation System Messages, and Tooling
Anthropic has launched Claude Opus 4.8, their most capable generally available model, featuring a 1M token context window and enhanced reasoning capabilities. The release also introduces mid-conversation system messages and publicly documents the `stop_details` field for refusal responses, benefiting developers integrating Claude into applications. Opus 4.8 is now available via Claude API, Amazon Bedrock, and Vertex AI.
- →Launch of Claude Opus 4.8 model
- →Mid-conversation system messages now available
- →Public documentation for `stop_details` field
- →Tooling and Claude Code enhancements for Opus 4.8
- →Updated default and performance parameters for Opus 4.8
Features (4) ›
- Launch of Claude Opus 4.8 model
The new Claude Opus 4.8 model is now generally available, offering a 1M token context window by default on the Claude API, Amazon Bedrock, and Vertex AI, with 128k max output tokens. It includes capability improvements, new features, and migration guidance detailed in its release notes.
- Mid-conversation system messages now available
Developers can now send `role: "system"` messages after a user turn on Claude Opus 4.8, preserving prompt cache hits when instructions change during long-running sessions. These messages are subject to placement rules and do not require a beta header.
- Public documentation for `stop_details` field
The `stop_details` field on refusal responses is now publicly documented, returning a `category` (cyber, bio, or null) and an explanation. This allows applications to route different classes of refusal to appropriate next steps.
- Tooling and Claude Code enhancements for Opus 4.8
The advisor tool, computer use tool, and task budgets now support Claude Opus 4.8. In Claude Code, Auto mode has been expanded, Max plan users default to fast mode on Opus 4.8, and Workflows are available as a research preview for multi-step agentic plans.
Enhancements (2) ›
- Updated default and performance parameters for Opus 4.8
On Claude Opus 4.8, the `effort` parameter defaults to `high`, and the minimum cacheable prompt length for prompt caching is reduced to 1,024 tokens. Adaptive thinking is enabled, triggering reasoning only when needed, reducing token usage compared to Opus 4.7.
- Opus 4.8 supports high-resolution image input and task budgets
Claude Opus 4.8 supports high-resolution image input up to 2576 pixels on the long edge, consistent with Claude Opus 4.7. Task budgets are also now supported for Claude Opus 4.8.
https://platform.claude.com/docs/en/release-notes/overview