Zed vs VS Code vs JetBrains: $10 Pro, $0 Core [2026] Zed's stable channel reached version 1.26.2 on October 28, 2026, the same day Microsoft released Visual Studio Code 1.144 with Copilot Fusion multi-model orchestration, AI image generation and experimental desktop computer control, while JetBrains made Air Governance generally available on October 22. Zed's release notes added syntax highlighting for Go 1.27 generic methods and AI-written commit messages, and dropped support for macOS Catalina and Big Sur, so the editor now requires macOS 12.0 (Monterey) or later. JetBrains also reached version 1.0 of its Java and Kotlin extension for VS Code and Cursor and reportedly added GitHub Copilot to Air on November 3 for local and cloud tasks. Zed vs VS Code stopped being a simple speed contest this fall. Microsoft shipped Visual Studio Code 1.144 on October 28, 2026 with Copilot Fusion multi-model orchestration, AI image generation and experimental desktop computer control https://visualstudiomagazine.com/articles/2026/10/28/vs-code-1-144-unleashes-copilot-fusion-ai-image-generation-and-computer-use.aspx . Zed’s stable channel reached 1.26.2 the same day, and JetBrains made Air Governance generally available on October 22. Three editors, three very different ideas about who should run your coding agents. JetBrains is hedging its bets, too. The company now sells its Java and Kotlin intelligence inside rival editors, and the JetBrains Java and Kotlin extension for VS Code and Cursor https://shattered.io/jetbrains-java-kotlin-vs-code-cursor-rc-2026/ has since reached version 1.0. That tells you where the market sits. A decade of picking one IDE and living inside it is giving way to mixed setups, where the editor, the agent and the billing meter come from different vendors. So the useful question in November 2026 is not which editor wins. It is where your agents run, who pays for the tokens, and how much of your day depends on a deep debugger and refactoring engine. Current versions, current prices and the numbers vendors have actually published answer that below, followed by a migration plan and a verdict. Don't miss new tech stories on Google Add Shattered once in the Google app and our stories show up in your news suggestions. Add now https://www.google.com/preferences/source?hl=en&q=https%3A%2F%2Fshattered.io%2F Where Zed, VS Code and JetBrains Stand on November 4, 2026 Start with the release state, because it shapes everything else. Zed ships weekly. Its newest stable build is 1.26.2, dated October 28, and a preview build numbered 1.27.0 carries the same date on Zed’s release page. The recent notes read like a working developer’s wish list: syntax highlighting for Go 1.27 generic methods, an option to open README files by default, and AI-written commit messages that follow the style of your repository’s recent commits. Zed also cut some old hardware loose. Its release notes removed support for macOS Catalina and Big Sur, so the editor now requires macOS 12.0 Monterey or later. Windows and Linux builds remain available. If your team still runs a fleet of aging MacBooks, that line item matters more than any benchmark. Visual Studio Code is on 1.144, released October 28, and the editor is free for private and commercial use. The headline change is less about the editor and more about agents. Microsoft made a new Copilot harness the default for new sessions in the editor and the Agents window, published version 1.0 of the open Agent Host Protocol, and moved a modernized interface with new Dark 2026 and Light 2026 themes from preview to the default, with a way back to the classic look during the transition. VS Code also keeps a few hard edges that matter on older machines. The last stable release to support 32-bit Windows is 1.83, and desktop builds on Linux need a distribution based on glibc 2.28 or later. Neither limit is new, but both explain why some corporate fleets lag behind the current build. JetBrains is the hardest to summarize because it is not one product. IntelliJ IDEA 2026.2.4 is the latest update on its blog, and posts about the 2026.3 line sound cues, remote development session management are already out. On top of the IDEs, JetBrains announced Air in October as a system of products for agentic development. Air Governance, the product formerly called JetBrains Central, became generally available on October 22 and gives organizations control over agents, models, providers and spending. One more recent item deserves a flag. JetBrains reportedly added GitHub Copilot https://shattered.io/github-copilot-fusion-ga-three-effort-tiers-30-percent-discount-2026/ to Air on November 3 for local and cloud tasks, with Copilot usage billed through the user’s Copilot plan rather than JetBrains AI credits. Cloud use reportedly needs Air Teams and administrator enablement. Treat that as a fresh launch whose details may still shift. No successor product hangs over any of the three. Zed’s preview channel is the closest thing to a next release, and it is a normal weekly step rather than a new generation. That makes this a clean moment to compare, since nobody is waiting on a major version that would rewrite the answer. Zed vs VS Code vs JetBrains: Full Specs Table The table below lines up twelve attributes using dated vendor material from the last few weeks. JetBrains cells stay general where the company ships many IDEs with different requirements. Where a vendor has not published a figure, the cell says so instead of guessing. | Spec | Zed | VS Code | JetBrains | |---|---|---|---| | Latest stable build | 1.26.2 Oct 28, 2026 | 1.144 Oct 28, 2026 | IntelliJ IDEA 2026.2.4 latest blog update | | Base price | Free Personal plan | Free, including commercial use | Paid IDE subscriptions, discounts for students, startups, nonprofits | | AI add-on price | Pro at $10/month | GitHub Copilot, free and paid plans | JetBrains AI plus Air plans | | Platform limits | macOS 12.0 or later, Windows, Linux | No 32-bit Windows after 1.83, Linux needs glibc 2.28 or later | System requirements vary by IDE | | Agent plumbing | External agents with session notices, context compaction | Agent Host Protocol 1.0, persistent sessions | Agent Client Protocol agents: Claude Agent, Codex, Gemini CLI, Junie | | Model access | BYOK for GPT-6.1 Sol, Bedrock lists GPT-6 Astra, Claude Fable 5.1, Claude Opus 5.5 | Copilot Fusion orchestrates multiple models | Junie, Mellum2.1, third-party agents | | Agent isolation | OS-enforced sandbox, default since 1.14 | Copilot harness default for new sessions | Air Governance controls access and spend | | Admin controls | Zed for Business, SSO, SAML and SCIM planned, not shipped | Organizational Copilot licensing | Air Governance, generally available Oct 22 | | Collaboration | Real-time collaboration built into the editor | Agent sessions continue across VS Code, GitHub apps, Copilot CLI | Air runs agents across IDEs, CLI, browsers, mobile | | In-house model | Zeta2.1 edit predictions | Copilot inline suggestions model | Mellum2 12B, open source , Mellum2.1 | | Recent interface change | Binaries 33% smaller on macOS | Modern UI on by default | 2026.3 adds sound cues, remote session management | | Student and discount offers | Student plan includes Pro features | Not applicable, editor is free | 40% off former students, 50% off universities | Read the table by row, not by column. Rows four through eight are where the editors diverge most, and they all concern agents. The base price row looks like a landslide for the free options, but it hides the fact that the price of a coding session now sits in the model tokens, not the editor license. Row five deserves a second look. Three protocols are competing to become the plug between an editor and an agent: Microsoft’s Agent Host Protocol, which just hit 1.0, JetBrains’ use of the Agent Client Protocol, and Zed’s support for external agents with session notices. If one of them becomes the default, the editor you pick today will matter less than the agent you plug into it. Choosing an editor that speaks an open protocol is the cheapest insurance against that shift. Speed, Memory and the Numbers Vendors Actually Publish Every Zed vs VS Code thread eventually argues about startup time and RAM. The honest answer is that none of the three vendors published a same-machine, same-repository measurement across all three editors this fall. What exists are vendor-reported efficiency figures, each measuring something different. They are worth reading, as long as you do not stack them into a ranking. | Data point | Value | Reported by | What it measures | |---|---|---|---| | Zed binary size, macOS | About 33% smaller | Zed release notes | Download and install footprint | | Zed binary size, Windows | About 28% smaller | Zed release notes | Download and install footprint | | Zed binary size, Linux | About 7% smaller 23% on the x86 64 release binary | Zed release notes | Download and install footprint | | Zeta2.1 edit predictions | 3x fewer tokens, 50 ms faster | Zed | Inline prediction cost and latency | | Copilot inline suggestions, configurable eagerness | 29% lower dismissal rate, 28% shorter suggestions vs medium | VS Code team blog | Suggestion quality at a chosen eagerness setting | | VS Code team focus | 50/50 split between AI and other features | VS Code Wrapped post | Engineering allocation | | Mellum2 size | 12 billion parameters, open source since June 1 | JetBrains | Model scale for coding workflows | Three vendors, three kinds of evidence. Zed’s numbers describe the cost of carrying and running the app, plus the cost of its prediction model. Microsoft’s numbers describe how developers react to suggestions. JetBrains’ figure describes the size of its own model. None of them tells you how fast a 400,000-line monorepo opens on your laptop. The Zeta2.1 figure is the most useful for budgeting. Fewer tokens per prediction should mean lower model spend, and Zed Pro bills overage at API list price plus 10%. If the savings carry over to hosted usage, a model that needs a third of the tokens makes that markup less painful. Microsoft’s eagerness data points a different way. It tells you that suggestion volume is a dial, and that the default setting is not always the best one for your team. Measure it yourself in 10 minutes. Your repository, your plugins and your OS decide the result, so run the check on your own machine. The script below launches one editor on a repository, waits 45 seconds for indexing and extensions to settle, then sums resident memory across every process whose name matches. Run it once per editor, on a clean login, with the same repository checked out at the same commit. bash /usr/bin/env bash usage: ./idle-rss.sh