AI Coding Tools Week 5: Claude Code Dominates, Releases Slow Down 27%

# AI Coding Tools Week 5: Claude Code Dominates, Releases Slow Down 27% ## TL;DR 📊 - Claude Code shipped 7 performance-focused updates, fixing startup issues and improving context management - Overall release velocity dropped 27% (19 vs 26 last week) as tools consolidate after holiday sprint ## This Week's Releases ### Claude Code (7 releases) The clear winner this week with versions 2.1.20-2.1.29. Major fixes include startup performance improvements for saved sessions, better debug logging for tool call failures, and context management validation fixes for gateway users. The `CLAUDECODEDISABLEEXPERIMENTALBETAS=1` flag now properly prevents validation errors. ### OpenAI Codex CLI (2 releases) Rust versions 0.92.0 and 0.93.0 brought SOCKS5 proxy support with policy enforcement and a new streaming plan mode with dedicated TUI. API v2 threads can now inject dynamic tools at startup and route calls through the server pipeline. ### Windsurf (3 releases) Versions 1.13.13, 1.13.14, and 1.9544.24 shipped with minimal feature details disclosed. ### Mystery Tool Updates (6 releases) Six rapid-fire releases (0.0.395-0.0.400) added MCP server instructions support, improved timeline displays for user responses, switched ordered lists to numbers, and added a theme picker. Compaction messages now show clearer command hints, plus Ctrl+X then / lets you run slash commands without losing input. ### Gemini CLI (1 release) Version 0.26.0 focused on PDF token estimation fixes, addressing a critical issue for document processing workflows. ## Highlight Deep Dive 🔍 Gemini CLI's PDF token estimation fix might seem minor, but it's actually huge for developers working with document AI. Accurate token counting is critical for cost management and context window planning when processing PDFs. This fix by @korade-krushna addresses a long-standing issue where token estimates were wildly off, leading to unexpected API costs and failed requests when documents exceeded context limits. For