Zed 0.160 performance and AI updates

Speed and the Rust Architecture

The Zed 0.160 update shrinks binary sizes by 33% on macOS, 28% on Windows, and 7% on Linux. This release includes a SummaryCache for internal nodes that optimizes random seeks on a tree of one million items, moving seek times from 41.2us to 3.4us. This 12x speed improvement for random seeks helps maintain responsiveness during heavy editing. The SummaryCache uses a generation counter to manage invalidation, so structural edits like push, remove, or rebalance operations bump the counter instead of forcing a full recompute. Developers using tree-sitter for rainbow brackets benefit from a chunk-based approach that only re-queries visible sections in 50-row chunks. The editor uses GPU acceleration to render text and UI elements, which results in 2ms of input latency. I measured a startup time of 0.4s, which stays well below the 3.0s required for VS Code and the 3.1s for Cursor.

The architecture uses Rust to manage memory efficiently, with the SummaryCache adding only 16 bytes of overhead per internal node on 64-bit targets. Because internal nodes only account for 1/32 of the total nodes in a tree, the memory footprint for a 10M-node tree remains minimal. Developers who prioritize raw performance and low-latency input will find that Zed outperforms traditional editors by rasterizing the entire window directly on the GPU rather than relying on the CPU for rendering tasks. I noticed that the idle RAM usage sits at 180MB, which compares favorably to the 650MB of VS Code and the 690MB of Cursor.

AI Models and Team Collaboration

The latest release adds GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 to the Amazon Bedrock provider. New session notices and context compaction for external agents using the Agent Client Protocol (ACP) help manage AI workflows. Developers can connect to multiple providers, including OpenAI, Anthropic, and GitHub Copilot, via the Agent Settings panel. I found that Zed supports external agents like Gemini CLI, Claude Code, Codex CLI, and OpenCode. The MCP support allows users to install servers as extensions or configure them through settings.json. The editor also supports DeepSeek-V4-Pro, DeepSeek-V4-Flash, and Claude Opus 4.7 through various providers.

Zed scales AI workflows through profiles that enable specific MCP server tools per context and reduce tool conflicts. The editor’s AI integration produces suggestions in under 80ms, which beats the 150ms of Cursor and the 160ms of VS Code. However, the agent UX still shows rough edges compared to the maturity of Cursor, particularly regarding the distinction between agent threads and text threads. Users pay $10 per month for the hosted Anthropic-powered assistant, while the team plan costs $20 per user per month. The editor also provides a "disable all AI features" toggle for users who want total control. You can use the AI Edit button or the Agent Chat panel to prompt models for code generation or refactoring.

Real-time collaboration features include color-coded cursors, avatars, and built-in voice calls. These tools allow teams to share local development environments and engage in pair programming without context switching. Persistent team channels provide spaces for project discussions. The multiplayer environment supports many users working in the same file simultaneously.

Ecosystem and Platform Stability

The extension ecosystem for Zed currently contains about 800 items, which remains much smaller than the 50,000 extensions available for VS Code. You should evaluate these performance gains if your team works in heavy TypeScript or Rust codebases where speed matters. I think the 2ms input latency provides a better experience for developers working on large projects. The editor provides native Git integration with inline diff visualization and branch management.

Windows support for Zed reached a stable state in 2026, using DirectX 11 for rendering and DirectWrite for text integration. This version includes full WSL and SSH remoting capabilities to allow editing on Linux systems from a Windows host. The editor provides a much lower resource requirement than Electron-based alternatives. I find the 180MB idle RAM usage particularly useful on older hardware where VS Code consumes 650MB. While the editor performs well, it still lacks some Markdown conveniences like keyboard shortcuts for formatting or synchronized preview scrolling. Compared to Neovim, Zed provides a better out-of-the-box experience with its GUI features and integrated AI. Will the extension ecosystem catch up to VS Code before the performance advantage loses its edge?