Skip to main content

Claude Code v2.1.280: Opus 5.5 Default, $4 Price Shakes Devs

Anthropic Drops a Bombshell: Claude Code v2.1.280

Remember those whispers about a secret internal test? Anthropic just confirmed them. In the latest release of Claude Code v2.1.280, Claude Opus 5.5 (claude-opus-5-5) officially takes the throne as the default Opus model. No more level-jumping rumors—this is the real deal. And the specs? A whopping 1 million token context window paired with a price tag that's almost unbelievable: $4 per million input tokens, $20 per million output tokens, and a cache read price of just $0.20 per million tokens. The leaked beta pricing was spot-on, which means top-tier intelligence is no longer a luxury—it's becoming an everyday tool.

Subscription Shake-Up: Sonnet Users Get a Surprise Upgrade

But wait, there's more. Anthropic didn't stop at pricing. They've quietly switched the default model for Pro and Team Standard plans from Sonnet to Opus. That's right—if you were comfortably using the cheaper tier, you've just been bumped up to the most expensive model. The billing curve? Let's just say it's heading north. However, Anthropic left a small escape hatch: effort levels saved before the /e effort model distinction won't apply to new models like Opus 5.5. You'll start at the default tier until you say otherwise. Opus 4.7, 4.8, and Fable 5 also got rule updates, no longer strictly following their original default effort levels.

The Little Fixes That Make a Big Difference

This update log reads like a small book, but most of it tackles those pesky edge cases that drive developers crazy. In full-screen mode, you can now scroll through the /skills list with your mouse wheel, and toggle specific skills in /plugin with a click. A new environment variable, CLAUDE_CODE_MAX_MCP_DESCRIPTION_LENGTH, gives you control over the 2048-character limit for MCP server tool descriptions. Security got a boost too: automatic mode no longer endlessly retries after a security check rejection—it simply says "retrying is futile." And when responses lag, it backs off and retries, stopping after ten attempts with a prompt.

Writing Chain Pitfalls? Fixed.

If you've ever had a Write call crash because the model included extra fields like path or file_text instead of the standard file_path and content, breathe easy. Validation no longer freaks out. Writing symbolic link paths now clearly tells you the real destination, and acceptEdits, allow rules, and automatic mode won't approve writes outside the tree. That frustrating API error—"role 'system' must precede an 'assistant' message"—won't kill every conversation. When the advisor tag isn't supported by the gateway, the request removes it and resends. MCP tool format errors no longer fail sessions during each round or when using /compact. Sub-agent caches are restored, rebuilding the original tool list instead of the re-sent one, which fixes the bug that destroyed prompt caches.

A Nod to Non-Latin Scripts

For users of Chinese, Persian, Arabic, and other non-Latin scripts, there's a thoughtful fix: the invisible character cleanup used to delete zero-width non-joiners, which are crucial for attaching suffixes (like the plural of "PDF") to Latin words or numbers. This time, they're spared. Windows terminal display issues after pressing Enter? Fixed with a screen redraw. Microphone secretly recording after Ctrl+C, Esc not canceling, spacebar triggering voice transcription in vim NORMAL mode—all resolved.

VS Code and Cloud Get Some Love

The editor now includes panels like /status, /sandbox, /chrome, /export, and /plan, and the /skills dialogue shows each skill's source, token estimate, and switch status. Pasted content over 800 characters or two line breaks gets marked, helping the model distinguish typed from pasted text. A startup deadlock from Python extension activation is resolved after 60 seconds, allowing independent operation. On the cloud side, GitHub Enterprise Server sessions now auto-renew tokens every eight hours, removing the "gh calls failing after eight hours" obstacle. Slack's Claude Tag gets a native Working indicator, Stop button, and thread title, plus notifications when guests join or leave channels.

Key Points

  • Opus 5.5 default: Now the standard for Pro and Team plans, with 1M token context.
  • Aggressive pricing: $4 per million input tokens, $20 output, $0.20 cache read.
  • Subscription shift: Sonnet users silently upgraded to Opus tier.
  • Edge-case fixes: Mouse scrolling, security retries, writing validation, and more.
  • Cross-platform improvements: VS Code panels, cloud token renewal, Slack integration.

From setting Opus 5.5 as default to quietly upgrading Sonnet users and fixing a laundry list of interaction issues, v2.1.280 sends a clear message: Anthropic isn't just making the model smarter—they're polishing the entire call chain so developers won't want to leave. With top-tier intelligence at $4 per million tokens and a million-token context, this battle for developers' wallets has reached others' living rooms.