This video compares Claude Opus 4.8 and Codex GPT 5.5 AI models in creating Blender 3D scenes, demonstrating that while both models can produce impressive results, their approaches and performance vary significantly across different complexity levels. Claude Opus 4.8 excels at detailed, stylized creations like the dragon (43 minutes), producing visually striking results with glowing accents and membrane wings, while Codex GPT 5.5 demonstrates superior efficiency and accuracy on complex technical scenes like the Starship rocket (10 minutes), achieving cleaner proportions and more realistic details. The comparison reveals that AI model performance depends heavily on task complexity, with simpler objects showing more consistent results across models, while complex scenes may favor different models based on their underlying architecture and optimization strategies.
Inmersión profunda
Prerrequisito
- No hay datos disponibles.
Próximos pasos
- No hay datos disponibles.
Inmersión profunda
Blender MCP: Who Won? Claude Opus 4.8 vs Codex GPT 5.5 AI Test and TutorialAñadido:
Who can create the best blender scenes?
GPT 5.5 or Claude Opus 4.8? In this video, we will have them start by creating a Pokéball, then we'll increase the difficulty by creating the Starship rocket from SpaceX, and last but not least, the final boss, the dragon. To get this to work, we first need to connect the Blender MCP. First, you need to run brew install UV in your terminal.
After you do that, you go to the Blender MCP GitHub repo and you download the add-on file. Then you open up Blender and click on edit. Then you have to click on preferences and after that, you have to click on add-ons. Click on install from disk, select the add-on file, and enable it. Then press N in the 3D viewport to open the side panel. Find Blender MCP and hit start MCP server.
Once you do that, Blender is ready for your AI agent to connect. For Claude code, you're working in the desktop app.
Go to settings, then developer, then edit config, and you'll see the Claude desktop config JSON file. You add a Blender entry under MCP servers with the command set to UVX and the argument Blender MCP. Save that, restart Claude, the Blender tools are loaded and ready.
For Codex, you navigate to settings, then click on the MCP servers. Click on add server, and this is what your configuration should look like. Click on save and reset the Codex app. Once it's configured, Codex picks up the Blender tools automatically when you start a session. Round one, the Pokéball. A simple shape, but every detail matters.
The proportions, the gloss, and that crisp red and white split. Claude running Opus 4.8 goes first. Watch Opus work. It blocks out the sphere, separates the shells, then adds the black band and the center button. Start to finish, this build took about 7 and 1/2 minutes. And here is Opus's Pokéball. The proportions are spot on, the button and chrome ring are clean, and the black band sits right where it should. Next to the reference, it is a strong match. The red skews a little coral and reads very glossy, almost candy-like, where the original is a flatter, brighter red. Opus actually flagged that itself, noting the red looked desaturated and pushing for a punchier tone. Overall, a faithful and polished first attempt. Now it is GPT 5.5's turn running through Codex. Same reference image, same task. Let's see how it stacks up. Codex takes a more deliberate path, building separate beveled shells, a recessed satin band, even procedural scuffs, and a full turntable animation. This one took about 12 and 1/2 minutes, noticeably slower than Opus. And the payoff is a clean, polished Pokéball sitting on a display plinth. The red is more muted and realistic, and the button has a subtle cyan glow. Against the reference, Codex nails the proportions, and that matte finish actually reads closer to the real toy. Two strong Pokéballs. Opus glossier and punchier, Codex more refined and restrained. The Pokéball is in the books. Time to raise the stakes with the Starship. Round two cranks up the difficulty. SpaceX's Starship on the launch tower. This one is all about scale, the lattice tower, the booster, and the venting plumes. Opus takes the first shot. Opus blocks out the tower and stacks the booster, but watch the chat. Renders timeout, and the connection keeps dropping. It fights through to a full model, though lighting trouble leaves its final shot dark. All in, this build dragged on to nearly 26 minutes. Here is what Opus put together before it gave out. The full tower and rocket are roughed in, but it errored before the render could resolve, so the whole shot stays dark and murky. By the way, this video was edited using Video Edit MCP, which is the only video editing platform built for the use of AI agents. The setup is simple. You just go to videoeditmcp.com, click on this curl command, and paste it into your terminal. Once it's done installing, the videoeditmcp app will open. Click your email, and click on connect AI agent. Then click connect.
The connection will show it's ready.
Once you open the AI agent, the app will finalize the connection. And that's it.
Your AI agent now has full video editing capabilities. Now, Codex steps up to the Starship. Same reference, same brutal complexity. Codex is methodical here.
The tower, the booster, the venting, even the Gateway to Mars text on the pad. And it just runs clean, finishing in about 10 minutes flat. Still, look at this result next to the reference. The proportions, the launch tower, the smoke, it's genuinely impressive. On Starship, Codex clearly pulls ahead.
Side-by-side, the silhouette and the venting really sell the shot. It's not flawless, the tower is simplified, but it captures the scene far better than Opus Manage. For the dragon, I went ahead and switched to Claude Code in the terminal. If you want to connect your Blender MCP to the Claude Code running on the terminal, you actually have to configure it separately from the desktop app. I've created this super terminal command to make it easier to connect it to Blender. I'll link it down below if you need it. You just copy and paste this entire command in the terminal, run Claude, and you're connected. Now, let's run this bad boy on the new Ultra Code.
Two builds down, one to go, and the dragon is the toughest test yet. Round three, the final boss. A fully detailed dragon, wings, horns, scales, and glowing accents, the works. And with Ultra Code driving Claude, I already know this is going to be good. Opus goes huge here, sculpting the body, extruding the neck and tail, then building out those massive membrane wings. And if you are enjoying this showdown, take a second to like and subscribe. It really does help. And this is Opus's dragon.
The glowing blue horns and spine accents are gorgeous, the wings are detailed, and the pose has real character. Against the reference, Opus captures the vibe beautifully. The icy glow, the membrane wings, the armored body. It is a little stylized and stockier than the sleek original, but seriously impressive. The catch? This was by far the slowest build of the day, clocking in at a massive 43 minutes. Opus even set up a turntable and studio lighting to show the dragon off from every angle. For the most ambitious build of the day, it's a fitting finale. Opus set an incredibly high bar, now all the pressure is on Codex. It works fast, blocking out the body, the long neck, the tail, and a big bat wing, all in about 12 and 1/2 minutes, less than a third of Opus's time. Here is the Codex dragon. I have to admit, I don't know what Codex was going for here. Put the images side-by-side, and I guess if you squint hard enough, they kind of look the same.
It's not good, but it's not horrible.
So, who do you guys think did better?
Comment your answer down below, and don't forget to like and subscribe.
Videos Relacionados
OpenHuman VS Hermes AI: Who Wins?
JulianGoldieSEO
285 views•2026-05-29
Long-Running Agents — Build an Agent That Never Forgets with Google ADK
suryakunju
142 views•2026-05-30
This computer is made from real human brain cells. And you can buy it.
Talktmsmedia
3K views•2026-05-28
BREAKING: Microsoft’s New Image Generating Model Beat Out GPT 1.5 and Nano Banana 2
aimmediahouse
122 views•2026-06-03
I Made the Same Anime Fight Scene in Every AI Video Generator
NobleGooseAnime
295 views•2026-05-30
Nvidia Bets Big On AI PCs | New Chip To Power Windows Laptops | Technology | AI Updates | N18S
cnnnews18
3K views•2026-06-01
I Tested NEW Opus 4.8 on Four Projects (Updated LLM Leaderboard)
AICodingDaily
298 views•2026-05-29
3D Platformer Update - NO CAPES
SolarLune
294 views•2026-05-30











