How to Use GPT-6 Astra to Control Blender with Codex

You can control Blender with GPT-6 Astra in Codex without learning every modeling tool first. Describe the object you want, and the official Computer Use plugin lets the agent inspect Blender and operate its interface. Your job is to refine the shape, check the result, and save the project at each milestone.

Choose how to access the model:

Your setup	Connection path
An account and subscription with access to Codex and Astra	Sign in to your official account and select GPT-6 Astra
No subscription	Connect a BetterToken API Key; official Codex account login is not required

Both paths use the same official Computer Use plugin installed from Plugins. The API Key changes model access and request billing; plugin installation stays the same.

This guide takes you from installation to an editable desk lamp and a short animation. You need a Mac or Windows computer capable of running Blender. Model requests travel over the internet, while Blender and rendering run on your computer.

From OpenAI’s official walkthrough. This is a separate architectural project; our exercise starts with a small lamp.

1. Install the app and connect the model in Codex

Visit the official ChatGPT download page, choose the desktop app for your system, and install it. In the current app, Codex is in the mode switcher. If you already have the standalone Codex App, update it and use its Codex section.

With a subscription

1. Launch the app and choose ChatGPT sign-in.
2. Sign in to the account that has the subscription.
3. Switch to Codex.
4. Select GPT-6 Astra in the model picker beside the message composer. If it is missing, check the models available to your account.

You do not need a separate BetterToken Key for this path. Continue to project creation.

Without a subscription: connect a BetterToken API Key

This path also works without an official Codex account. You use your own BetterToken account and pay for model requests by usage.

Create a BetterToken API Key to use GPT-6 Astra in Codex for Blender.

If Codex already uses an API Key

Replace the Base URL and API Key in your existing API connection with BetterToken’s details. You do not need to run the installer again.

1. Create a key for the GPT provider in BetterToken.
2. Open your current Codex API provider settings. Set Base URL (base_url) to https://www.bettertoken.ai/v1 and replace the existing API Key with your own BetterToken key. If the key is stored in an environment variable, update that value.
3. Confirm the selected model is gpt-6-astra. Change the model ID if you previously used another model.
4. Save, fully restart Codex, and send a short message to test the connection.

Once it works, continue to the project and Blender. Reuse Computer Use if it is already installed; otherwise install it using the steps below.

If this is your first API connection

1. Open Workspace in BetterToken and create an API Key for the GPT provider. Copy it for setup on your computer.
2. In the key connection setup, choose Codex and gpt-6-astra. Use the full model ID; do not shorten it to astra.
3. Open the BetterToken Codex guide and its Command-line setup section. The installer needs Node.js; install it first as directed if it is missing.
4. Fully quit Codex. Open Terminal on Mac or PowerShell on Windows and run the command for your system below.
5. Paste your key locally when the installer requests it. Do not send it to the agent in chat. Wait for setup and the connection check to finish.

macOS — Terminal:

curl -fsSL "https://www.bettertoken.ai/install-codex-provider.sh" -o /tmp/install-codex-provider.sh
bash /tmp/install-codex-provider.sh --model gpt-6-astra

Windows — PowerShell:

iwr "https://www.bettertoken.ai/install-codex-provider.ps1" -OutFile "$env:TEMP\install-codex-provider.ps1"
powershell -ExecutionPolicy Bypass -File "$env:TEMP\install-codex-provider.ps1" -Model "gpt-6-astra"

The installer configures the BetterToken provider and Responses API connection, then makes a short model test. That test also consumes API tokens. Reopen the app and select Codex after setup finishes. ChatGPT sign-in is not required for this connection path.

If setup fails, resolve the error first: check the key, model access, and funds for the API request. Installing Computer Use will not fix a model connection error.

2. Create a local project

1. Create an astra-blender folder on your computer, for example inside Documents.
2. In Codex, click the + beside Projects and add this folder as a local project. Labels may vary by language and app version; choose a project on your computer.
3. Start a new task inside it and confirm Astra is selected.
4. Send a short message: “Reply with one word: ready.”

A reply without a connection error means you can move on to Blender. With an API connection, the request should also appear in BetterToken’s usage history. This checks the model; we will test access to the Blender window separately.

Keep the .blend file, images, and animation in this project folder. Return to the same task for later changes so Codex can see your earlier instructions.

3. Install and open Blender

Blender is free software for 3D modeling, materials, animation, and rendering. Download it from the official Blender page.

1. Choose the build for your system. On Mac, check whether you have Apple Silicon or Intel.
2. On macOS, open the downloaded disk image and drag Blender to Applications. On Windows, run the installer and follow its steps.
3. Launch Blender. You can keep the initial defaults and choose a new General file.
4. You will usually see a starter cube in the center and Cube, Camera, and Light in the scene list on the right. Leave them in place for now.
5. Use File → Save As to save desk-lamp.blend in astra-blender.

To ask Codex to help install Blender, send this in your project:

Help me install Blender from https://www.blender.org/download/ for this computer.
Check whether Blender is already installed first. Use the official stable build.
Explain any installer or system-permission step I need to complete myself.
Open Blender and tell me which version is running.

Installation is complete when the Blender window is open. A download link in Codex’s reply is not enough. Platform instructions are in the Blender manual.

4. Install official Computer Use

These steps are the same for subscriptions and BetterToken API Keys:

1. Open Plugins in the app.
2. Find Computer Use by OpenAI and click Install plugin.
3. If Enable appears after installation, enable it.
4. In the plugin card, enable its MCP server and Skill. The server supplies control tools; the Skill explains how to use them.
5. Open Settings → Computer use and enable Any App control if that switch appears in your version. Grant access to Blender when requested.
6. On Mac, grant the requested Screen Recording and Accessibility permissions in system settings. Restart the app if prompted.
7. Return to the project task. If necessary, start a new task to load the installed plugin’s tools.

The official Computer Use guide shows where these settings are. On Windows, keep Blender visible on the active desktop and leave the computer unlocked while the agent works: it uses that session’s mouse and keyboard.

Now test actual interface control:

Use the official Computer Use plugin to inspect the open Blender window.
Tell me the current file name and the objects visible in the scene list.
Through the Blender interface, rename Cube to Connection_Test.
Show the updated scene list, then stop. Do not create or delete objects.

Look at Blender: the cube should stay in place, with its name changed to Connection_Test in the list. If Codex only describes the actions, ask it to perform them through Computer Use. This small test confirms that it can see and change the right window.

5. Build your first model

Start with something easy to inspect: a desk lamp with a round base, stem, and angled shade. For the first pass, ask Codex to use the interface so you can follow its actions.

Use Computer Use to build a small desk lamp in the current Blender project.
Work through Blender's graphical interface for this first pass. Do not generate
the geometry with a Python script or command-line process.

Replace only the Connection_Test object from our connection check.
Create a round base, a vertical stem, a short angled arm, a conical shade,
and a bulb. Keep the parts as separate editable objects named Lamp_Base,
Lamp_Stem, Lamp_Arm, Lamp_Shade, and Lamp_Bulb.

Use simple proportions and place the lamp upright on a flat surface.
Set a camera that shows the entire lamp. Save desk-lamp.blend in this project.
Show a front view and a side view, then stop for my feedback.

Check the silhouette first: the base rests on the surface, the stem connects to the shade, and the whole lamp fits in frame. Select a part in the object list. If the base and shade can be selected separately, the model remains editable.

Fix one visible problem at a time: “make the base wider,” “shorten the stem,” or “tilt the shade toward the desk.” Save after each successful change so you can return to an accepted shape if the next edit goes wrong.

Geometry from OpenAI’s official project. This view helps inspect objects and intersections independently of attractive lighting. It illustrates the approach, not the output of the lamp exercise.

6. Add materials and lighting

Once the shape works, refine its appearance:

Continue with the existing desk lamp. Keep its geometry and object names.
Give the base and shade a muted blue painted-metal material, the stem a brushed
metal material, and the bulb a warm glow. Add soft studio lighting and a neutral
background. Do not add decorations or rebuild the lamp.

Render a 640x360 preview from the camera. Inspect the image, fix obvious floating
parts, intersections, or framing problems, and save preview-front.png and
desk-lamp.blend in this project. Stop after the preview for my feedback.

The agent may suggest Blender’s Python API, bpy, for repetitive operations. Repeat your interface-only constraint if you want to keep working that way. If faster scene changes matter more, you can allow code or add the official MCP below. In either case, ask it to open and visually inspect the result.

Materials and lighting in OpenAI’s official example. Astra combined bpy with render inspection in that project; this image does not demonstrate a mouse-only workflow.

For later work: add official Blender MCP

Computer Use operates the interface. MCP gives Codex a separate connection to the running Blender instance for reading the scene and performing operations. Add it when manual actions become numerous; the first exercise above does not require it.

1. Open Blender’s MCP page and follow the official Setup instructions.
2. Check your Blender version: the current official project needs 5.1 or later.
3. Give Codex the Setup link in the same task and ask it to install the Blender extension and the separate MCP server, then register the server in Codex. These are the two parts of the connection.
4. Once the extension is prepared, open Edit → Preferences → Add-ons → Install from Disk in Blender, choose its ZIP, and enable it. Codex can guide you through this with Computer Use.
5. Start the connection server in the extension settings. The extension and MCP server must use matching addresses and ports; use the values in the current official instructions.
6. Start a new task in the same Codex project and ask MCP to read the object list. Compare Lamp_Base, Lamp_Shade, and the other parts with Blender.

Do not start editing until the object lists match. You can then use MCP for construction and Computer Use to inspect the window and rendered frame. If the connection drops, check the scene’s current state before creating parts again.

7. Make a short animation

Save the successful static model first. A camera move around it is enough for a first clip:

Use the existing lamp scene to create a 6-second camera orbit at 24 fps,
frames 1 through 144. Keep the lamp still and all parts editable.
Move the camera smoothly through a quarter-circle around the lamp, keeping the
whole object in frame. Do not change the materials or add new objects to the lamp.

Save a separate file named desk-lamp-animation.blend.
Render 640x360 preview images for frames 1, 72, and 144 as
orbit-001.png, orbit-072.png, and orbit-144.png.
Inspect those images and stop before rendering the full animation.

Open all three images. The camera should not pass through the lamp, crop the shade, or show an empty frame. If they look right, ask for all 144 frames to be rendered into a separate folder and assembled as desk-lamp-orbit.mp4. Keep the first video at 640×360; higher resolution increases local rendering time.

If rendering is slow, ask for the current render’s progress. Do not start a second render of the same scene until you know whether the first has finished.

Troubleshooting

What you see	What to check
Codex does not reply, returns 401, or cannot find the model	For API access, check the key, GPT provider, gpt-6-astra, and setup result; fully restart the app
The model replies but Computer Use tools are missing	Check plugin installation and the server and Skill switches; start a new task
Tools exist but the window is inaccessible	Grant Blender access; check Screen Recording and Accessibility on Mac, or an active unlocked desktop on Windows
The agent describes steps but changes nothing	Explicitly ask for a small Computer Use action and the resulting state
MCP connects but shows another scene	Check the open .blend file, Blender instance, and extension address/port
A call is interrupted or times out	Request the current objects and saved files first, then resume from the last completed stage
The image looks wrong	Return to the shape and one viewpoint; fix proportions, intersections, and lighting before a full render

You should finish with desk-lamp.blend, a separate animation file, inspection images, and—after the full render—a video. Reopen the .blend file and change one part yourself, then continue the project with a specific edit for Codex.