Claude Code v2.1.280: Opus 5.5 becomes the default Opus model

Claude Code v2.1.280, released September 22, is the first Claude Code release to ship with Claude Opus 5.5. Anthropic now makes claude-opus-5-5 the default Opus model in Claude Code.

Opus 5.5 in Claude Code

Anthropic documents Opus 5.5 with a 1M-token context window, $4/Mtok input, $20/Mtok output, and $0.20/Mtok cache reads. Anthropic's model announcement says it costs about 40% less than Opus 5 on typical workloads and generates output more than 30% faster. Fast mode is also available for Opus 5.5 where supported.

For teams that explicitly pin model IDs, this is not a reason to silently rewrite production configuration. Test the new model and effort behavior against your own coding and automation evals before changing pinned workloads.

MCP and observability

v2.1.280 adds CLAUDE_CODE_MAX_MCP_DESCRIPTION_LENGTH, allowing operators to override the default 2,048-character cap applied to MCP tool descriptions and server instructions across the session.

The hook_execution_complete OpenTelemetry event now reports hook output sizes and how many oversized outputs were written to a file. That should make it easier to diagnose hooks that unexpectedly create large outputs or add operational overhead.

Permission and automation fixes

The release includes several fixes that matter for unattended and managed environments:

• writes through symlinked paths are judged by where the write actually lands, preventing an in-tree spelling from implicitly approving an out-of-tree destination;
• auto mode stops repeatedly retrying actions rejected by a safety check and backs off when the classifier returns no answer;
• conversations behind gateways that do not support the advisor tag retry without it rather than failing every turn;
• resumed sessions no longer start an unsolicited model turn merely because unfinished background work exists;
• messages sent to a background subagent in headless and SDK sessions are less likely to be lost as that agent finishes;
• long overloads no longer leave cloud or self-hosted runner sessions permanently unauthenticated after access-token rotation.

These are mostly reliability fixes rather than new APIs, but they are relevant if Claude Code is part of CI, a self-hosted runner fleet, or long-running agent automation.

Interactive and plugin improvements

Mouse support now covers more fullscreen lists, including /skills and plugin state options. The release also fixes several dialog/keybinding issues, plugin update metadata, private marketplace credential handling, MCP re-authentication state, and LSP use from background subagents.

Anthropic also changed the default model on Pro and Team Standard plans from Sonnet to Opus, matching Max, Team Premium, and Enterprise. Previously saved effort choices no longer automatically carry over to newly released models such as Opus 5.5; new models begin at their own default until you select an effort level.

What to do

For interactive users, update and verify the installed version:

claude update
claude --version

For production automation, test v2.1.280 against your gateway, MCP servers, permission rules, hooks, background agents, and any explicit model or effort configuration before broad rollout.

Official Claude Code changelog · Claude Code releases · Claude Opus 5.5 announcement · documentation index