Zed performance dominates as JetBrains replaces Fleet with Air

Zed outperforms competitors in speed and efficiency

Zed outperforms VS Code in speed and resource efficiency. In April 2026, benchmarks showed Zed cold-starting in 0.60 seconds compared to VS Code’s 1.29 seconds. A clean install without cache widened this to 0.40 seconds for Zed versus 3.00 seconds for VS Code. In May 2026, ToolChew found Zed v1.2.6 cold-started in 0.12 seconds. This 16x memory gap is massive. VS Code spawns 23 processes using 3,549 MB of RAM for a folder, whereas Zed uses 5 processes and 222 MB. Even with AI features active, Zed uses 340 MB while VS Code reaches 980 MB. Large file handling favors Zed; it opens 100,000-line files 8x faster than VS Code, which takes 1.19 seconds to render the same file in 0.15 seconds. Input latency remains a differentiator. Zed maintains 2ms latency, while VS Code reaches 12ms or 25ms. One user on Pop!_OS reported an annoying 1 second delay in navigation despite having all dependencies. You probably understand how Electron bloat impacts your hardware. Zed uses its custom GPUI framework to talk to the GPU via Metal on macOS or DirectX on Windows.

JetBrains replaces Fleet with agentic environments

JetBrains ended Fleet on December 22, 2025. The company replaced the tool with Air, an agentic environment where users supervise AI agents instead of writing code. Air works with Anthropic’s Claude and remains macOS-only, though Windows and Linux support arrives in 2026. This move follows the departure of Fleet from the market after it failed to differentiate itself from IntelliJ or VS Code. JetBrains now works with Zed on the Agent Client Protocol (ACP) to standardize how agents connect to editors. This protocol draws an analogy to the Language Server Protocol. OpenAI also developed the Codex App Server to decouple agent logic from client surfaces. Using JSON-RPC streamed as JSONL over stdio, the bidirectional protocol allows for conversation primitives like Items, Turns, and Threads. One user request unfolds into a sequence of actions including user input, agent progress, and produced artifacts. This architecture handles streaming diffs and approval flows where the server pauses the turn to wait for an "allow" or "deny" response. One deployment pattern uses a worker to provision a container and launch the App Server inside it. Why does JetBrains continue to move away from traditional IDE structures?

Collaboration and pricing for 2026 teams

Zed provides built-in collaboration tools that VS Code requires extensions like Live Share to use. Users add contacts via GitHub usernames and share projects directly. They track collaborators using cursors and avatars or use built-in chat and audio channels. Zed also includes channels for team discussions. This functionality exists out-of-the-box in Zed. While VS Code users rely on a marketplace of over 100,000 plugins to manage their workflows, Zed relies on its WebAssembly-based extension model to achieve high performance without the heavy overhead of a browser-based runtime. Zed’s Pro plan includes Zed-hosted models and unlimited edit predictions. A Business plan provides an admin layer with org-wide controls over AI features and unified billing. Business users can bring their own API keys from Anthropic, OpenAI, or Google. The Pro plan allows for monthly billing, while Business users can contact sales for order form contracts at 25+ seats. Users can access a 14-day free trial of Pro which includes $5 of GPT-5.6 Luna usage.