Latest stable Zed: 1.25.3, released (release notes). 390 stable releases since 3 Jan 2024.
Zedrelease notes, changelog and setup guides
952builds since 3 Jan 2024
390stable releases since Jan 2024
2days between stable releases, median
25in May 2025, the busiest month
20days, the longest wait (4 Jan 2024 to 24 Jan 2024)
2024202520261.0
390 stable562 previewBusiest week: 20 builds from 16 Feb 2026
Zed has shipped one stable minor a week since 1.0 on 29 April 2026. That is 1.0 through 1.25, and 54 stable builds counting patches.
Zed says v1.25 (21 October) cut the binary by about 30% on every platform. The patch after it, v1.25.3 on 23 October, updated dependencies "to address recent software ecosystem vulnerabilities". The notes don't say which ones.
As of 24 October 2026.
01Latest stable builds
1.25.3Updated dependencies to address recent software ecosystem vulnerabilities.
1.25.2Added GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 to the Amazon Bedrock provider.
1.24.4Fixed OpenCode models failing to load with a missing field npm error.
1.24.3Added Gemini 4 Argon to the Google AI models
1.24.2Improved Agent Panel terminal support by allowing Terminal tabs to be dragged into it.
1.23.3Fixed an issue where no fonts would show for the markdown preview configurations in the settings UI.
1.23.2Added the agent.maxidleretainedthreads setting to control how many idle agent threads remain loaded.
1.22.0Added an optional model parameter to spawnagent to choose a model for that spawn instead of using the configured subagent model or inheriting the parent…
1.21.0Added SuperGrok sign-in so SuperGrok subscribers can use Grok models in the Agent Panel.
1.20.2Fixed payment errors from non-Zed model providers incorrectly prompting users to upgrade to Zed Pro instead of displaying the original provider error message.
Zed started training its own model. Edit prediction with Zeta arrived on 13 February 2025, and Zeta was rebuilt as Zeta2 in March 2026.
Since 4 February 2026 you choose the edit prediction provider yourself. Copilot is still in there: 70 stable releases since 2024 mention it, the latest v1.21.0 on 23 September 2026. On 1 June 2026 Zed explained what GitHub's usage-based Copilot billing means for its users.
Collaboration turned into a product of its own. DeltaDB was introduced on 11 June 2026 and Delta on 12 August. Since 16 September Delta is in public beta, pitched as a replacement for pull requests.
Windows took fifteen more months. On 19 August 2025 the blog asked "What's Taking So Long?!" and on 15 October 2025 it answered with "Windows When? Windows Now".
The REPL stayed. It comes up in the notes of 17 stable releases since 2024, most recently v1.8.2 on 24 June 2026, which added a Stop button to notebook cells. Setup is in the REPL section below.
Billing became the recurring topic. Usage-based billing for heavy users came on 17 October 2024 and token-based pricing on 24 September 2025. In May and June 2026 two more posts covered Anthropic's and GitHub's billing changes.
On 23 July 2025 Zed let you switch all AI features off. On 5 May 2026 it published "We're Not Building AI Features for the Money".
Yes, GitHub Copilot works in Zed 1.25, but it is not the default. Zed ships its own model, Zeta, and Copilot is one of six edit prediction providers you pick with the edit_predictions.provider setting.
Copilot plugs in at three points: inline edit predictions you accept with tab, GitHub Copilot Chat as a model provider for the Agent Panel, the inline assistant and commit messages, and a Copilot external agent where one is available. It has worked in Zed since v0.82 in April 2023.
Make Copilot the edit prediction provider in your settings file:
{
"edit_predictions": { "provider": "copilot" }
}
Click the Copilot icon in the status bar. Zed shows a device code: copy it, click “Connect to GitHub” and paste the code on GitHub's verification page. The popup closes once you are authorized.
For Copilot Chat, run agent: open settings, open LLM Providers and sign in to GitHub Copilot Chat. Then pick a Copilot model from the model selector in an Agent Panel thread; Zed saves it as agent.default_model.
On Copilot Enterprise, point Zed at your instance. The setting covers both edit predictions and Copilot Chat, and sign-in then redirects to your enterprise URL:
The six edit prediction providers in the Zed docs (October 2026), with their default debounce:
zed, the default: Zeta, Zed's open-source model. 0 ms. Needs Pro, the Pro trial, Student or Business.
copilot: GitHub Copilot. 75 ms. Needs a Copilot sign-in.
codestral: Mistral Codestral. 150 ms. Needs a Codestral API key.
mercury: Mercury Coder from Inception Labs. 0 ms. Needs an Inception API key.
ollama: any local Ollama model. 0 ms.
open_ai_compatible_api: vLLM, llama.cpp server, LocalAI or another /v1/completions server. 0 ms.
On Zed's Free plan the documented prediction options are Copilot or a local model. Change the debounce with prediction_debounce under the provider's own settings. "provider": "none" stops predictions only; "disable_ai": true switches off every AI feature, Copilot included.
Zed's post of 1 June 2026 covers GitHub's switch to token-based billing: Copilot Chat use in Zed, including agent turns, inline assists and commit messages, is metered in GitHub AI Credits. Copilot edit predictions are not.
Fixes since 1.0: v1.3.5 (20 May 2026) stopped Copilot predictions from depending on a system Node install. v1.16.2 (24 August) fixed Copilot Chat sign-in and routing for GitHub Enterprise Cloud. v1.21.0 (23 September) fixed auto-compaction for Copilot models whose prompt limit is below their context window.
70 stable release notes since 2024 mention GitHub Copilot, the latest 1.21.0 on 23 September 2026.
Zed's REPL, introduced on 24 July 2024, runs Jupyter kernels inside the editor and prints results under the code that produced them. The documented kernels are Python (ipykernel), TypeScript (Deno), R (Ark or Xeus), Julia and Scala (Almond).
If you added the kernel while Zed was open, run repl: refresh kernelspecs. repl: sessions lists what Zed found.
Open a file in that language and run repl: run (ctrl-shift-enter on macOS) on a line, a selection or a block.
For notebook-style work, split a script into cells with # %% in Python or // %% in TypeScript. repl: run then runs one cell at a time.
On macOS the system Python does not work as a kernel; the docs say to use pyenv or a virtual environment. A default kernel per language goes under jupyter.kernel_selections in settings.
The REPL docs describe notebooks as scripts with cell markers. They do not describe opening .ipynb files, and no stable release note since 2024 mentions ipynb.
17 stable release notes since 2024 mention REPL, the latest 1.8.2 on 24 June 2026.
GPUI is the Rust UI framework Zed is built on, developed in the zed-industries/zed repository. Its site calls it “a fast, hybrid immediate and retained mode, GPU-accelerated UI framework for Rust”.
Zed moved to a rewritten GPUI 2 on the preview channel on 3 January 2024. Nathan Sobo's post covers what changed, and Thorsten Ball's “Why the big rewrite?” from 23 January 2024 covers why. Zed Weekly #27 on 24 November 2023 described the design GPUI 2 settled on for views and elements.
On crates.io, gpui sat at 0.1.0 from June 2022 until 0.2.0 on 9 October 2025. The latest is 0.2.2 from 22 October 2025. The newest GPUI line in a stable Zed release note is from v1.18.0 on 2 September 2026: better text input for GPUI web applications on Android.
6 stable release notes since 2024 mention GPUI, the latest 1.18.0 on 2 September 2026.
Mostly on Wednesdays. 159 of 390 stable builds from 3 January 2024 to 23 October 2026 were published on a Wednesday (UTC), more than Monday, Tuesday and Thursday combined. Across all 952 builds the busiest hour starts at 18:00 UTC.
Every Zed build, stable and preview, dropped into a weekday by hour grid in release order (10 second loop). Brighter cells hold more builds; the bars count builds per weekday: Mon 118, Tue 129, Wed 340, Thu 172, Fri 163, Sat 20, Sun 10. Made by dev-news from the zed-industries/zed GitHub releases.
07Questions people search for
What is the latest version of Zed?
Zed 1.25.3, released 23 October 2026 (release notes). 54 stable builds have shipped since 1.0 on 29 April 2026.
How do I add GitHub Copilot to Zed?
Set "edit_predictions": { "provider": "copilot" } in your settings, click the Copilot icon in the status bar and confirm the device code on GitHub. For Copilot Chat models in the Agent Panel, sign in under agent: open settings > LLM Providers.
Does Zed support GitHub Copilot Enterprise?
Yes. Set enterprise_uri under copilot in your settings to the URL your GitHub Enterprise admin gives you. It applies to Copilot edit predictions and Copilot Chat, and sign-in then goes through your enterprise instance.
Is Copilot the default AI in Zed?
No. The default edit prediction provider is Zeta, Zed's own open-source model, included with Pro, the Pro trial, Student and Business. Copilot is one of six providers, set with edit_predictions.provider.
What can I use in Zed instead of Copilot?
For predictions: Zeta (zed), Codestral, Mercury Coder, Ollama or any OpenAI-compatible completions server. For the Agent Panel: Zed-hosted models, a ChatGPT subscription, your own API keys, local models, or an external agent such as Claude Agent or Codex.
Can Zed run Jupyter notebooks?
Zed runs Jupyter kernels in its built-in REPL and treats scripts with # %% cell markers as notebooks. Its docs do not describe opening .ipynb files.
What is GPUI?
GPUI is the GPU-accelerated Rust UI framework Zed is built with. It lives in the zed-industries/zed repository and is published on crates.io as gpui.
08From @zeddotdev on X
Zed@zeddotdev
🚀 Zed v1.25 is out! Since v1.23, we've reduced Zed's binary size by ~30% across all platforms.
The size cut was announced a week ahead of the release.
Zed@zeddotdev
Developers built Google Docs and Figma so others wouldn't lose work, then ironically kept using decades-old tools that let us lose our own. @conradirwin explains how DeltaDB uses CRDTs to sync every keystroke so no work is lost.
