Claude Code 2.1.280 Makes Opus 5.5 the Default Opus Model

Claude Code 2.1.280 changes the default Opus selection to Claude Opus 5.5. The release notes list a 1M-token context window and pricing of $4 per million input tokens and $20 per million output tokens, with cache reads priced at $0.20 per million tokens. The public npm registry independently reports 2.1.280 as the package's latest version.

Claude Opus 5.5 becomes the default in Claude Code

The default change matters for users who select the Opus tier without pinning a model identifier. Existing workflows that depend on Opus now resolve to claude-opus-5-5, according to the release notes. The 1M context window gives large repositories, long agent sessions, and broad code-review tasks more room before context management becomes a constraint.

The pricing figures in the release note are a usage qualifier, not a claim that every Claude Code subscription is billed directly at those rates. Teams using API-backed or enterprise arrangements should check the billing path and any account-level limits that apply to their deployment.

What else ships in version 2.1.280

The release also adds CLAUDE_CODE_MAX_MCP_DESCRIPTION_LENGTH, allowing users to change the 2,048-character cap applied to MCP tool descriptions and server instructions for a session. That is useful for tool-heavy environments where the default metadata limit clips important operating detail.

Several safety and reliability fixes affect automated workflows. Symlinked writes are now judged by their actual landing path when Claude evaluates prompts, edit rules, and auto mode. Auto mode also stops retrying after a declined safety review and backs off when a safety check returns no answer. The notes additionally cover prompt validation, background subagents, plugin updates, dialog handling, voice dictation, Windows terminals, and long-running sessions.

Access, verification, and migration notes

The npm registry's latest tag points to 2.1.280, providing an independent check that the package release is available through the public distribution channel. The GitHub release remains the authoritative source for the feature and fix list. Users who need reproducibility should pin their Claude Code version and explicitly configure the model where a workflow cannot tolerate a default-model change.

Before rolling the update across a team, review cost controls, MCP metadata size, and any tests that assume a particular Opus model. The practical change is straightforward: upgrading Claude Code can now change which Opus generation an unpinned workflow uses, while the wider fix set improves the surrounding agent and terminal experience.

Sources and methodology

This article reports the official Claude Code release notes and cross-checks package availability against the public npm registry. It does not infer subscription entitlements from API token pricing, and it does not treat the release's long fix list as evidence of a separate product launch.