A post on social media got a lot of attention: someone kept using the same Astra model, switched from Codex to Zed, and said everything suddenly felt quicker. Their usage meter also seemed to drop much more slowly.
If you're wondering what Zed actually is, the confusion is understandable. Zed, GitHub, Astra and Codex all sound like they belong to the same product, and they don't.
Short answer: Zed is not a GitHub replacement and not an Astra replacement. It's the app where you open your code, tell an AI agent what to do, edit files, run a terminal and check the changes. The AI model can stay exactly the same, but if the setup around it changes, the experience can change a lot.
1. The 30-second answer
Once you split these four apart, it gets easy.
| Name | What it does, roughly |
|---|---|
| Zed | The editor and workbench where you open code and work with an AI agent |
| GitHub | The online home for your code, its history, branches and pull requests |
| GPT-6 Astra | The AI model that reasons and writes code or text |
| Codex | OpenAI's coding-agent tool, a separate way to get software work done with AI |
Here's a handy picture: GitHub is the warehouse and the ledger, Zed is the workbench, Astra is the brain, and Codex is another assembly line.
And it isn't an either-or choice between Zed and Codex. Zed's own docs describe three ways to work: Zed's built-in agent, Codex as an external agent connected through ACP (the Agent Client Protocol), and Codex CLI running in Zed's terminal.[1][2][3]
2. What is Zed IDE?
Zed is a code editor, in the same family as VS Code and Cursor. You'll often see it called Zed Editor.
It's built in Rust and draws its interface with the GPU. Zed 1.0 was announced in April 2026 and runs on macOS, Linux and Windows.[4]
These days Zed has AI built in deeply. With Zed Agent, the AI can search your project, read files, rewrite code, run commands in the terminal and show you the changes to review.[2]
So think less "fancy notepad" and more "a coding desk with an AI helper, a terminal and Git all sitting on it."
3. Zed, GitHub, Astra and Codex, one at a time
The most common mix-up is treating Zed and GitHub as the same kind of thing.
GitHub stores your repository online, keeps the history, and handles branches and pull requests. Zed is the app on your own computer where you open that repository and work on it.
Astra is something else again. OpenAI announced GPT-6 Astra in September 2026, an AI model that handles software engineering, computer use, browsing, research and more.[5]
Codex is easier to understand as a product for running software work through an agent, rather than as the name of a model. OpenAI's own help pages present Chat, Work and Codex as separate places to work inside ChatGPT.[6]
So you can use the same Astra and still choose which route to call it from.
4. What does "moving Astra from Codex to Zed" actually change?
The Astra model doesn't move onto your computer, of course.
What changes is the machinery that hands context to the AI, reads and edits files, runs tools and the terminal, shows diffs and keeps the conversation history. People often call this the agent "harness." Zed's docs also say the route that starts, displays and controls an agent is sometimes called a harness.[2]
In Zed there are at least three routes:
- Use OpenAI models with Zed Agent.
- Run Codex inside Zed as an external agent through ACP.
- Run Codex CLI in Zed's Terminal Threads or an ordinary terminal.
That's why "I moved from Codex to Zed" doesn't tell you exactly which setup someone is using.
The brain stayed the same. What changed is the body and the cockpit it sits in.
5. Why would the same AI feel faster?
The time an AI task takes isn't only the model thinking.
Roughly speaking:
Total time
= model reasoning
+ gathering context
+ searching and reading files
+ tool calls
+ running the terminal
+ retries
+ waiting on connections
+ showing and reviewing diffs
Even with the same model, the later items can change. Say someone kept hitting reconnect waits and retries in one setup and barely meets them in another. The model thinks at the same speed, yet it can still feel like a different tool.
That said, Zed isn't guaranteed to be faster. Searching a huge repository broadly every time makes preparing context heavy, and putting an external agent in between adds its own work.
So how fast it feels depends on the whole setup, not just the name of the model.
6. If usage barely drops, does that mean unlimited?
This part is easy to get wrong.
In May 2026, Zed announced that you can sign in to Zed Agent with your ChatGPT account and use OpenAI models under the same ChatGPT usage allowance you get with Codex. You don't need a separate API key, and it's a different route from pay-per-use API billing.[1][7]
But that doesn't mean limits disappear once you install Zed. The official explanation says usage depends on your ChatGPT plan.[1]
If your usage seems not to be dropping, there are several possible reasons:
- Each job needs less context or fewer tool round trips
- There are fewer retries
- The usage display doesn't match the internal processing units
- The tasks you compared simply weren't the same size
So it's too early to conclude "Zed means unlimited" from a feeling alone.
Tokens haven't become immortal.
7. If I move to Zed, don't I need Codex anymore?
Neither "yes" nor "no" fits well, because you can use Codex itself inside Zed.[2][3]
Zed Agent is a natural pick when you want tight editor integration for file edits, code search, terminal commands, MCP and so on. If you want Codex's own agent behavior and settings, you can run it in Zed as an ACP external agent, or run Codex CLI in the terminal.
Astra also has features specific to Codex. OpenAI says that on long jobs that run past the context window, Astra in Codex uses a new way of keeping earlier context in a searchable form.[5]
So switching to Zed doesn't make Codex worthless.
You can even mix them: like Zed as your editor, but still want Codex as your agent.
8. How does this compare to working through GitHub?
If you already let AI fix things in a GitHub-centered project, this will feel familiar.
The difference is that GitHub is the remote side where your work is saved and merged, not the desk you work at.
A typical flow looks like this:
GitHub repository
↓ clone / pull
Open the project in Zed
↓
Give instructions to the AI agent
↓
Search and edit files
↓
Run tests in the terminal
↓
Check the diff
↓
commit / push
↓
History, PR and CI on GitHub
If so far you felt like the AI was working directly inside GitHub, Zed is more like bringing the same GitHub repository onto your own computer and working on it with the AI at your workbench.
Same repo, different workspace.
9. What a real workday looks like
Say you're fixing a bug in an existing service. You open Zed and then the repository. Then you ask the AI: "Find the cause, fix it, run the related tests and explain the diff."
With Zed Agent, the AI can find and fix the code using the editor's own tools. If you'd rather have Codex's behavior, open Codex as an external agent or start the CLI in the terminal.[2][3]
After the AI's fix, check the diff and test results, commit with Git and push to GitHub. On a team, you continue with a pull request and CI.
The key point: adopting Zed doesn't make GitHub go away. Your local work in Zed simply connects to the history, review and merging on GitHub.
10. Which route should you pick?
Rather than asking "which is the best," it's easier to choose by what you want to keep.
| What you want | Route that fits |
|---|---|
| Let the AI fix code mainly through Zed's screens, editing and diff review | Zed Agent |
| Keep Zed's interface but use Codex itself as the agent | Codex via ACP |
| Keep Codex CLI's settings and feel exactly as they are | Codex CLI / Terminal Threads |
| Use broader computer tasks beyond chat, or OpenAI's other work surfaces | Use ChatGPT / Work / Codex alongside |
| Store, share and review your code online with its history | Keep using GitHub |
The point is that choosing Zed doesn't mean dropping Codex.
11. Common misunderstandings
"Zed replaces GitHub."
It doesn't. Zed is the workbench; GitHub is where the repository, its history and teamwork live online.
"Astra itself gets faster in Zed."
Not necessarily. What changes is the whole experience, including the harness, connections, how context is built and how tools run.
"Zed means unlimited tokens."
The official information doesn't say that. With a ChatGPT subscription, the usage allowance of your plan applies.[1]
"Zed and Codex are rivals, so I can only use one."
You can use both. Zed can run Codex as an ACP external agent, and Codex CLI also runs in the terminal.[2][3]
"Same model, so changing the environment makes no difference."
It does. Even with the same model, how context is passed, how many tools are used, retries, connection stability and the review flow all change the total time.
12. Summary
Zed IDE is neither an AI model nor a GitHub replacement. It's a fast working environment where you open code and edit, run and check things together with an AI.
When someone says they "moved Astra from Codex to Zed," what really changed is less the brain and more the route that makes the brain work.
So the same Astra feeling faster is entirely possible. But "it felt faster" doesn't prove the model got faster, and "my usage didn't drop" doesn't prove it's unlimited.
And the best part is that this isn't a simple Zed-versus-Codex showdown.
You can seat Codex at Zed's desk.
Bring the same repository from the GitHub warehouse, let Astra work on it at Zed's workbench, use Codex's assembly line when it helps, and send it back to GitHub when you're done.
Once you pull the buzzwords apart, it's a pretty ordinary story of warehouse, workbench, brain and assembly line.
References (7)
- Zed, “Use Your ChatGPT Subscription in Zed zed.dev
- Zed Docs, “Agents zed.dev
- Zed Docs, “External Agents zed.dev
- Zed, “Zed 1.0 zed.dev
- OpenAI, “GPT-6 Astra: A new generation of intelligence openai.com
- OpenAI Help Center, “ChatGPT Work and Codex help.openai.com
- Zed Docs, “Use an Existing Subscription zed.dev
