← Back
Anthropic
Claude Code 2.1.70 fixes API compatibility, model responses, and Windows audio issues
Claude Code · releasebugfixfeatureperformanceintegration · github.com ↗

Bug Fixes

This release resolves several high-impact issues affecting API compatibility and core functionality:

  • Third-party gateway support: Fixed API 400 errors when using ANTHROPIC_BASE_URL with third-party gateways by improving proxy detection for tool reference blocks
  • Bedrock compatibility: Resolved "This model does not support the effort parameter" errors when using custom Bedrock inference profiles or non-standard Claude model identifiers
  • Model response handling: Fixed empty model responses immediately after ToolSearch by correcting system-prompt-style tag rendering at the prompt tail
  • Windows audio: Fixed voice mode failures on Windows native binary and push-to-talk activation issues
  • SSH performance: Fixed Enter key inserting newlines instead of submitting on slow SSH connections
  • Text handling: Corrected clipboard corruption for non-ASCII text (CJK, emoji) on Windows/WSL using PowerShell Set-Clipboard

New Features

VS Code Integration Enhancements:

  • Added spark icon in activity bar listing all Claude Code sessions, with sessions opening as full editors
  • Added full markdown document view for plans with comment support for feedback
  • Added native MCP server management dialog accessible via /mcp command for enabling/disabling servers and managing OAuth authentication

Performance Improvements

  • Reduced prompt input re-renders by ~74% during turns
  • Reduced startup memory by ~426KB for users without custom CA certificates
  • Cut Remote Control polling from 1–2s to once per 10 minutes while connected, reducing server load ~300×
  • Improved compaction by preserving images in the summarizer request for faster, cheaper prompt cache reuse

Other Improvements

  • Enhanced error messages for microphone silence detection
  • /rename command now works while Claude is processing
  • Skill listing injection reduced by ~600 tokens per resume
  • Fixed repeated model update notifications for legacy Opus versions