Claude Code v2.1.284 adds Sonnet 5.5, spend tracking in USD, and multiple UX fixes

Anthropic released Claude Code v2.1.284, which sets Claude Sonnet 5.5 as the default model for the Anthropic API. The model offers a 1M context window and new pricing at $2 per million tokens for input, $10 per million tokens for output, and $0.20 per million tokens for cache reads.

The update adds dollar-denominated spend tracking to the /usage command and status line, displaying monthly totals such as “$271.40 / $500.00 spent this month” when the gateway runs this version or later. The status line’s rate_limits.spend_limit field now includes used_usd, limit_usd, and period fields.

New keybinding actions—effortSlider:decreaseEffort, increaseEffort, and toggleUltracode—allow users to rebind the /effort slider’s arrow and Tab keys via keybindings.json.

The /rate-limit-options command is now documented in /help and the command menu for claude.ai subscribers, providing a discoverable way to access usage-limit notices.

A new /mcp reconnect all command in the interactive terminal retries all MCP servers that failed to connect or need authentication at once.

The release includes multiple stability and correctness fixes: damaged response streams no longer show raw errors or the word “undefined”; session recovery after server errors is improved; “Prompt is too long” errors are retried after compaction; and model-unavailable notices now include a Learn more link.

Additional fixes address malformed image sources in Agent SDK sessions, MCP tool calls in resumed sessions, repeated calls to plan-usage endpoints after rate limits, and incorrect reporting when MCP servers are restricted by managed settings.

Terminal rendering issues are resolved, including fullscreen output erasure, scroll position jumps, dialog tab bar truncation, and vim mode repeat commands failing over SSH or in tmux.