Claude Code v2.1.284 Adds Sonnet 5.5 Default, /mcp reconnect all, and Gateway Spend Tracking
Claude Code v2.1.284 is out, and the headline change is a model swap: Claude Sonnet 5.5 (claude-sonnet-5-5) is now the default Sonnet model on the Anthropic API. It runs a 1M-token context window at $2/$10 per Mtok with $0.20/Mtok cache reads.
New commands and keybindings
/mcp reconnect all— retries every MCP server that failed to connect or needs authentication, in one shot, from the interactive terminal.effortSlider:decreaseEffort,effortSlider:increaseEffort, andeffortSlider:toggleUltracode— new keybinding actions so the/effortslider's arrow and Tab keys can be rebound inkeybindings.json./rate-limit-options— now listed in/helpand the command menu for claude.ai subscribers, so the usage-limit notices that mention it point at a command you can actually find.
Gateway spend tracking
When the Claude apps gateway runs v2.1.284 or later, /usage and the status line now show dollar amounts against the spend limit — e.g. "$271.40 / $500.00 spent this month". The status line's rate_limits.spend_limit also gains used_usd, limit_usd, and period fields.
Gateway admins also get startup warnings when a managed policy's availableModels is empty, or omits the model Claude Code starts on without setting model or enforceAvailableModels. Two auth additions: auth: { google: {} } for telemetry.forward_to destinations (exports telemetry to Google Cloud's OTLP endpoint using the gateway's own GCP credentials), and certificate client authentication via private_key_jwt for IdPs that issue certificate credentials instead of client secrets.
Bug fixes worth noting
- Damaged response streams no longer surface raw errors like
"JSON Parse error"or write the literal word "undefined" into an answer — they're retried or reported as interrupted. - Overloaded/server errors arriving right after a thinking block are now retried instead of ending the turn.
- "Prompt is too long" errors that survived compacting now trigger one more compact pass, keeping less of the recent conversation.
- Agent SDK sessions no longer crash on a user message with a malformed image
source; the malformed image is replaced with an explanatory note. - MCP tool calls in a resumed session no longer fail with "No such tool available" while the server is connecting — the call now waits up to 10 seconds.
claude mcp addrefuses (with guidance) when managed settings restrict MCP servers to plugins, instead of reporting false success.- The
/pluginconfigure screen now uses a true/false choice for booleans, rejects invalid numbers, and uses ←/→ to change an options field. ANTHROPIC_FOUNDRY_RESOURCEis validated before being interpolated into the Foundry endpoint host.- On Windows with many plugins, nonexistent plugin
bin/directories are no longer added to PATH. sparsePathsmarketplaces no longer clone empty on git before 2.39 and wipe a working local copy.
If you're on a Claude apps gateway, upgrading is what unlocks the USD spend display and the 1M-context marking for Claude Desktop.
📖 Read the full source: GitHub Claude-Code
👀 See Also
Amazon Employees 'Tokenmaxxing' with MeshClaw AI Agents to Meet Usage Targets
Amazon developers are automating unnecessary tasks with the internal MeshClaw tool to inflate AI token consumption, after the company set weekly usage targets for 80% of devs and introduced internal leaderboards.

OpenClaw's New Release: A Simple Name Change or a Major Upgrade?
OpenClaw, previously known as ClawDBot, has undergone a transformation. Read on to find out whether this change is merely cosmetic or introduces new features and improved stability.

LibreOffice Online Development Resumes After Community Vote
The Document Foundation has resumed work on LibreOffice Online after a community vote nullified the 2022 freeze. TDF will reopen the repository for contributions but won't host servers—instead providing self-hostable tools.
Debian Votes on AI/LLM Contribution Policy: What Developers Need to Know
Debian has begun voting on the future of AI/LLM contributions. The outcome will shape how AI-generated code is handled in Debian packages.