Update: Claude Sonnet 5.5 ships as Claude Code's default, $2/$10

Version 2.1.284 also frees Ultracode from effort, surfaces spend in real dollars, and adds one-command MCP reconnect — all in the same drop as the model.

Nowline SEP 28 11:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • The rumor is now the default

    Claude Code 2.1.284 makes claude-sonnet-5-5 the default Sonnet, with a 1M-token context window at $2/$10 per Mtok and $0.20/Mtok cache reads — the grey-tested model reported this morning, now shipped. Your everyday Sonnet just went from a 200K to a 1M window.

  • Ultracode unhooks from effort

    Ultracode now toggles independently and no longer forces xhigh effort, with new /effort keybindings to increase, decrease, or toggle it. You can crank parallel agent throughput without automatically paying the top-tier reasoning cost.

  • Spend limits, in actual dollars

    /usage and the status line now show real amounts (e.g. "$271.40 / $500.00 spent this month"), and rate_limits.spend_limit gains used_usd, limit_usd, and period. Budget tracking stops being a guess from percentages.

  • One command to un-stick MCP

    /mcp reconnect all retries every failed MCP server at once, and MCP servers can now use URL-mode elicitation for browser-based auth flows. No more reconnecting dropped servers one at a time.

  • Auto mode's classifier moves server-side

    The permission classifier now runs server-side by default (local only when telemetry is off), and read-only or sandboxed shell commands wait for server review. Opt out with CLAUDE_CODE_AUTO_MODE_SERVER=0.