Zed Review: The AI Editor That Made Cursor Feel Sluggish

What it’s actually good at

The speed is the thing you notice in the first ten minutes and then can never unsee. Built with Rust and its own GPU UI stack, Zed renders screens at roughly 120 FPS, turning searches and scrolls that take seconds in Electron-based IDEs into millisecond operations. Opening a 100k-line file? Instant. Fuzzy-searching across a monorepo? Instant. Going back to Cursor after a week in Zed feels like typing through a wet paper towel.

The AI story is where Zed gets interesting in 2026, and it isn’t the chat panel, it’s the open Agent Client Protocol. Zed ships an in-editor agent panel, edit predictions, and MCP tool support of its own, but the feature that sets it apart is ACP, an open standard that lets an external coding agent (Claude Code, OpenAI’s Codex CLI, Gemini CLI) drive the editor directly, with diffs rendered inline in Zed’s UI rather than dumped into a terminal pane. That’s the most forward-thinking thing anyone’s shipped for the agent-lock-in problem. You aren’t buying into Zed’s private agent, you’re bringing your own, and Zed just gives it a faster home.

Edit predictions are the day-to-day workhorse. Zed ships its own autocomplete model (Zeta), a full agent mode with multi-thread support, 15 LLM providers, and as of May 20, 2026, Terminal Threads that let you run Claude Code or Amp as a sidebar agent alongside your existing editor context. Zeta doesn’t just predict the next token, it predicts the next edit, so accepting a multi-line change is one tab press. It isn’t quite Cursor Tab yet, but it’s close, and it feels local because the editor around it is so responsive.

The pricing is the other reason to pay attention. Personal, $0 forever: full editor, 2,000 accepted edit predictions per month, unlimited use with your own API keys or external agents such as Claude Agent and Codex CLI, unlimited local models via Ollama, LM Studio or llama.cpp, full real-time collaboration, macOS/Linux/Windows. Zed-hosted models require a paid plan. Pro, $10/month: unlimited edit predictions, access to all Zed-hosted models, $5 of tokens included each month, usage-based billing beyond $5 with a default $10/month incremental spend limit. Two-week free trial with $20 of token credits, Claude Opus excluded from the trial. Business, $30/seat/month: unlimited edit predictions, org-wide AI model policies, data governance controls, role-based access controls, and unified spend visibility with seats and AI usage on one invoice. That’s not a “come for the trial, leave when the credits die” pricing page. The free tier is genuinely useful for daily work if you have an API key of your own, and Pro is half the price of the market leader.

Real-time collaboration is the sleeper feature. Zed includes multiplayer editing with human teammates as a core feature, not a bolted-on extension: shared cursors, voice chat, and true co-editing in the same file. I tested this pairing on a bug fix across two machines, and the experience was closer to Google Docs than to screen-sharing. Both of us could edit different parts of the same file at once with no lag or merge friction. This is a category Cursor simply doesn’t compete in, its collaborative features are AI-facing (chat, agent sessions), not human-to-human. If your team does a lot of remote pair programming, this alone is worth the switch.

Where it lets you down

The extension ecosystem is the honest tradeoff. It is not a VS Code fork. That matters for two reasons: first, it means the extension shelf (500+ as of today, versus VS Code’s 50,000+) is genuinely smaller. Second, it means the editing engine (keybindings, cursor movement, multi-cursor, buffer management) was designed without VS Code constraints. The second half of that is arguably a good thing, Zed’s multi-cursor and multibuffer implementations are cleaner than VS Code’s, but the first half will bite you. If you rely on a specific linter, formatter, or DSL extension that only exists for VS Code, check before you switch.

The agent depth is the other place Cursor still has an edge. Cursor wins on the deepest codebase features: Composer multi-file edits, @codebase retrieval, Background Agents running autonomous tasks, and a codebase index that’s been optimized over three major versions. Zed’s Agent Panel plus ACP external agents covers most of the same ground, but Cursor’s out-of-the-box retrieval is still stronger on very large repos. In practice, this matters most on the “rewrite this whole subsystem” class of task. For everything else, writing new code, fixing bugs, refactoring a file or three, Zed’s agent panel, or Claude Code running through ACP, does the job.

And then there’s the credit trap on Pro, which is the one thing worth flagging clearly. Pro at $10/month includes unlimited Zeta predictions (the autocomplete model runs on Zed’s infrastructure) and $5 in LLM token credits. The $5 credit sounds like a rounding error, and it is. A single afternoon of moderate Claude Sonnet usage in Agent mode will clear it. Once you exceed $5, you’re billed at the model provider’s list price plus a 10% markup, with invoices triggered monthly or per additional $10 spent, whichever comes first. The default spend cap is $10/month, which is fair, but if you’re going to lean on Zed-hosted Agent mode all day, you’re better off bringing your own Anthropic or OpenAI key and treating Pro as an autocomplete subscription.

Should you pay for it?

Depends on what you already have. If you’re brand new to AI editors and want the deepest built-in experience, Cursor Pro at $20/month is still the safer buy. The Composer/Tab combo is more polished for the “just tell me what to install” user. But if you already pay for Claude Code, or you have an OpenAI API key, or you’re the kind of developer who cares about editor performance the way audiophiles care about DACs, Zed is the better deal by a wide margin.

The path I’d actually recommend: install Personal, run it for a week with your own API key or Claude Code piped in through ACP, and see if the speed and the workflow click. If they do, $10/month for Pro gets you unlimited edit predictions and you’re done. If you’re on a team doing remote pair programming, Business at $30/seat/month for the admin controls is honest pricing for what you get.

The bottom line

Zed delivers on its core promise: it’s fast in a way that changes how it feels to use an editor all day, and it treats “bring your own AI agent” as a first-class idea rather than an afterthought. ACP is the most forward-thinking approach we’ve seen to the agent-vendor-lock-in problem, and native real-time collaboration is a real, useful feature that most competitors skip entirely. It isn’t trying to out-build Cursor’s Composer, and it shouldn’t be judged as if it were. If you’re already happy running Claude Code or Codex CLI from the terminal and want a faster, more legible home for that workflow, or if raw speed and native collaboration matter more to you than the deepest possible built-in AI, Zed is one of the best editors available right now. It’s the pick for the developer who wants to own their agent stack, and at $10/month for Pro, it earns its keep on the first day.