Disclosure: some links on this page may be affiliate links; we may earn a commission at no extra cost to you. Read the disclosure.

By the toolist.ai editors. Researched from Zed's own pages and hands-on tests published by others, not hands-on tested by toolist.ai. Updated October 21, 2026.

Zed review: pricing, features and what testers found

Zed is an open-source code editor written in Rust with its own GPU-drawn interface instead of Electron, with a built-in AI agent, edit predictions and room for outside agents such as Claude Agent and Codex. The editor is free with no account, and its AI runs on your own API keys, subscriptions or agents; Pro costs $10 a month for Zed-hosted models with $5 of tokens included and unlimited edit predictions.

Our take

Zed suits developers who want a fast, native code editor, written in Rust rather than built on Electron like VS Code, and who want to choose their own AI: the open-source editor is free with no account, and its agent panel runs your own API keys, a ChatGPT or Copilot subscription, local models or outside agents such as Claude Agent and Codex, at no charge from Zed. Pro costs $10 a month for Zed-hosted models with $5 of tokens included, then provider list price plus 10%, and unlimited Zeta edit predictions, which the free plan lost on October 7, 2026; Business costs $30 a seat a month for admin controls, with no included AI credit. XDA's Anurag Singh saw a small project take under 100MB of RAM in Zed against more than 1GB in VS Code (January 2026), and by August 2026, having replaced VS Code with Zed, ran Codex, Claude and Zed's own agent side by side in its Threads Sidebar, though he found VS Code still ahead on extensions. Testing builds from before version 1.0 in late 2025, InfoWorld's Serdar Yegulalp found the editor comfortable and leaner than VS Code but Zeta's predictions uneven and prone to breaking Python indentation, and The New Stack's David Eastman, on Windows 10, showed multibuffers editing many files at once but hit Windows rough edges such as a file with a byte-order mark it did not recognize. The catches are a small $5 token allowance with usage billed past it up to a spend limit (by default $10 more a month), payments that are not refundable, Business without SSO or SOC 2 yet, and tests that mostly predate version 1.0 (April 29, 2026). This verdict rests on those four published tests and Zed's own pages.

Based on Zed's published plans and the hands-on tests linked below, not on a test of our own.

Good for

  • Developers who find VS Code or its forks heavy and want a fast native editor, bringing their VS Code or Cursor settings and a familiar keymap with them
  • Running several coding agents side by side (Zed's own, Claude Agent, Codex, Gemini CLI and others) on subscriptions or API keys you already pay for, with nothing extra to pay Zed
  • Developers who want AI kept optional: the editor works without an account, and one setting (disable_ai) turns off every AI feature, including the agent panel, the Threads Sidebar and edit predictions
  • Light users of hosted models who want one $10 bill with $5 of tokens and unlimited edit predictions, and verified university students aged 18 or over, who get a free year of the Student plan: Pro's features with $10 a month of token credits (usage stops there), but without Claude Fable, Claude Opus, GPT-6 Astra or the GPT pro models
  • Teams that want per-seat admin control over hosted AI, data sharing and spend, can bring their own Anthropic, OpenAI, Google or AWS contracts, and can do without SSO for now

Watch out for

  • Pro includes only $5 of tokens a month, counted at Zed's prices (provider list price plus 10%, from $0.055 per million input tokens on GPT-5 nano to $198 per million output tokens on GPT-5.5 pro and GPT-5.4 pro). Past it, usage bills up to a monthly spend limit that defaults to $10 more (so $20 in all; you can set it to $0), with mid-month invoices for each $10 of extra usage, a threshold Zed may raise to as much as $100, and hosted models stop once the limit is reached. A failed payment also blocks hosted models.
  • Since October 7, 2026 the free plan no longer includes Zed's hosted Zeta edit predictions, which it had offered as 2,000 accepted predictions (InfoWorld's 2025 test still describes them); Zeta now needs Pro, the trial, Student or Business, and free users must switch to another provider such as GitHub Copilot or a local model. Zed's pull request 65258, merged that day, states the change and updates the editor's messaging; we found no blog post announcing it.
  • Fees are in US dollars, with tax added, and are non-refundable except where the law requires; self-serve plans bill monthly only (Zed publishes no yearly price; Business contracts of 25 or more seats go through sales), and subscriptions renew automatically unless you cancel before the renewal date. Business has no free trial and no included AI credit, and its org-wide spend limit starts at $0, so an admin must raise it before members can use hosted models. The Pro trial lasts 14 days or until its $5 of GPT-6 Luna (the only hosted model in the trial) runs out, needs no card and turns back into Free on its own; that $5 is shared with Delta, Zed Industries' separate app for coding with agents and teammates.
  • Zed charges nothing for external agents, but they bill through their own providers and can change: Anthropic announced in May 2026 that Claude Code used through ACP would draw on a separate Agent SDK credit rather than Claude Pro or Max limits, then told subscribers the change was not taking effect yet (Zed's update of June 16, 2026), and GitHub moved Copilot Chat, including Copilot Chat models used in Zed, to token-based billing on June 1, 2026.
  • Zed says it keeps no prompts or code by default, and its contracts bar hosted-model providers from training on them and require zero data retention, except for Anthropic's Covered Models (including Claude Fable 5.1 and Claude Fable 5), whose prompts and outputs Anthropic keeps for at least 30 days for safety review. Rating an AI response sends the whole thread to Zed, edit-prediction training data is collected only if you opt in and only from open-source projects, client-side usage metrics and crash reports can be turned off in settings, and server-side telemetry is collected whenever you use hosted AI or collaboration. With your own keys, subscriptions or external agents, the provider's own terms apply.
  • Business has no SSO, SAML, SCIM or SOC 2 yet: Zed says it is working on them, and its SOC 2 page says it is working toward Type 1. Zed's AI service is for users aged 18 or over, and its terms include binding arbitration with a 30-day opt-out. The agent sandbox covers only Zed Agent's terminal and fetch tools, not external agents, Terminal Threads, language servers or your own terminal, and on Windows it needs WSL and may not stop every escape.
  • Three of the four cited tests predate version 1.0 (April 29, 2026): InfoWorld's (version 0.212.4) and The New Stack's from late 2025, and XDA's January 2026 speed comparison, all before Parallel Agents, Terminal Threads, the Business plan and the end of free Zeta predictions. Only XDA's August 2026 piece covers Zed since 1.0, still before the end of free Zeta predictions, and both XDA pieces are one writer's accounts of personal use. None of the four tried Zed-hosted models on Pro or measured what the $5 of tokens covers.

What Zed does

Price

Free plan: Personal (Zed Free) is the whole open-source editor, free forever and usable without signing in. Its AI runs only on what you bring: your own API keys (Anthropic, OpenAI, Google, Amazon Bedrock, OpenRouter, DeepSeek, Mistral and more), a ChatGPT Plus or Pro or GitHub Copilot subscription, local models through Ollama or LM Studio, or external agents such as Claude Agent and Codex, each billed by its own provider. Zed-hosted models need Pro, and since October 7, 2026 Free no longer includes Zed's own Zeta edit predictions (it had 2,000 accepted predictions); other prediction providers such as GitHub Copilot or a local model still work. Real-time collaboration needs a GitHub sign-in.

PlanBilled monthlyBilled yearly, per monthIncludes
Personal (Free) free not listed The full open-source editor with no account needed; AI through your own API keys, ChatGPT or GitHub Copilot subscriptions, local models and external agents such as Claude Agent and Codex, billed by their providers; edit predictions only from other providers such as GitHub Copilot or a local model (Zed's hosted Zeta predictions ended on Free on October 7, 2026); no Zed-hosted models; collaboration with a GitHub sign-in
Pro $10 not listed Zed-hosted models from Anthropic, OpenAI and Google with $5 of token credit a month, then usage at provider list price plus 10% up to a monthly spend limit (by default $10 more); unlimited Zeta edit predictions; a 14-day free trial with $5 of GPT-6 Luna and no card needed. Billed monthly from the day you subscribe
Business $30 not listed Price per seat, self-serve with no minimum, month to month; members get Zed-hosted models and unlimited edit predictions, with no AI credit included (hosted usage at provider list price plus 10%, or your own keys at no extra charge); org-wide controls over hosted models, edit predictions, collaboration and data sharing, an org spend limit, roles and permissions, and one invoice for seats and AI use; no free trial; order-form contracts and invoice billing from 25 seats through sales; no SSO, SCIM or SOC 2 yet

Prices last checked: 2026-10-30 on Zed's pricing page. Full breakdown: Zed pricing.

Price history

No price change since we first recorded these prices on October 21, 2026. We re-check them every week and list each change here.

What hands-on testers found

We have not tested Zed ourselves. These are the published tests we found that used it hands-on, newest first; follow the links for the full reviews.

XDA · Anurag Singh · August 4, 2026

What they tested: Zed as his main editor in place of VS Code, on a MacBook with 16GB of RAM, after the Parallel Agents update (April 2026) and Terminal Threads (May 2026), running Codex, Claude and Zed's own agent in separate threads. He names no version, plan or model, and did not cover Zed-hosted models, edit predictions or prices.

  • He noticed Zed opening larger projects faster than VS Code, which in his account gets sluggish and takes 3 to 4GB of RAM once extensions pile up; he quotes Zed's own figures for opening a large project (0.25 to 0.6 seconds against VS Code's 1.3 to 3.8) but says he could not verify them.
  • Parallel Agents is what let him leave VS Code: in the Threads Sidebar he could run Codex on one project, Claude on another and Zed's native agent beside them, keep each thread to chosen folders or its own Git worktree, and run terminal agents alongside ACP agents.
  • He found VS Code still ahead on its extension library, a more mature Live Share, advanced debugging, Git and remote tooling and deeper Copilot integration, but had not met a VS Code extension he could not replace in Zed; he still calls VS Code the best option for teams that want stability.

XDA · Anurag Singh · January 11, 2026

What they tested: Zed against VS Code and Google Antigravity for speed and memory use, in January 2026, before version 1.0 (no version named). He reports no use of Zed's agent, a paid plan or Zed-hosted models.

  • In his use, one small project needed under 100MB of RAM in Zed but over 1GB in VS Code, and a large codebase took a few seconds to open in Zed against 10 to 15 seconds in VS Code.
  • Opening projects felt instant almost every time, and file navigation, search and autocomplete ran with no delay he could perceive, without stutters or dropped frames while background tasks ran.
  • He counted VS Code's extension marketplace as its biggest remaining advantage, since Zed cannot run VS Code extensions and uses its own WebAssembly extensions instead.

InfoWorld · Serdar Yegulalp · updated November 19, 2025

What they tested: Version 0.212.4, which he treated as pre-release, writing a sample Python application (a simple wiki that renders Markdown to HTML with templates) with Zeta edit predictions on, and looked at setup, settings import, collaboration channels and extensions. First published November 17, 2025 (the page's metadata). It predates version 1.0, Parallel Agents, Business and the October 2026 end of free Zeta predictions, and he reports no use of the agent or a paid plan.

  • Zeta's predictions were uneven: some accurate, useful snippets for boilerplate code but a hindrance elsewhere, so he turned them off once the routine parts were written, and they often got indentation wrong, which matters a lot in Python.
  • He could import VS Code or Cursor settings and pick a keymap from a familiar editor, though some shortcuts still differed (word wrap, for one), and the native GPU-drawn interface resized and redrew smoothly, even with text files of several megabytes.
  • Extensions were numerous and covered even niche needs, but could not be switched off per project, only uninstalled; he concluded Zed was comfortable, fast and leaner than web-based editors such as VS Code, and worth trying.

The New Stack · David Eastman · October 28, 2025

What they tested: Zed's new Windows release on his old Windows 10 machine in October 2025: a fresh install with Sublime Text key bindings, a C# project, multibuffers and multiple cursors, signing in to Claude as the agent and the built-in terminal. He chose not to test the AI itself, and the test predates version 1.0 and today's plans.

  • He showed multibuffers at work: a search for references showed every match as an editable window into its file on one page, and multiple cursors changed a class name in all of them at once.
  • He hit Windows rough edges: a C# file saved as UTF-8 with a byte-order mark, which Zed did not recognize, showed a stray first character, diagnostics did not refresh smoothly, and the welcome page's Clone Repository option failed without explanation and was gone after an update.
  • He uses Zed as his 'other' editor beside Visual Studio and VS Code for C#, expects it to gain ground in that role, and thinks its emphasis on speed and AI puts it a little out of step with many developers.

Test result

toolist.ai has not run its own test of Zed; the findings above come from the published tests linked. See how we research and test.

Best coding tools · Zed pricing

Do you make Zed? Claim this listing to correct it; it's free. To see whether ChatGPT, Claude, Gemini and Perplexity recommend it, and how many readers we send you: toolist.ai for AI companies.

Sources