Opus 5.5 vs Fable 5.2 – Building the Taj Mahal in Blender

I wanted to see how two Claude models compare at generating 3D content, so I gave both the same job in Blender: build the Taj Mahal to scale, and give me a file I can 3D print. Claude drove Blender 5.3 directly through the official Blender connector. I did no modeling myself.

Opus 5.5: the first attempt

I started with Opus 5.5. My brief was one line: a hyper-realistic model of the Taj Mahal, to scale, textured and 3D printable.

About 11 minutes later I had a Blender scene in real-world metres and a print-ready STL. Opus looked up the main dimensions before modeling and wrote the whole model as Python scripts that it ran inside Blender, checking the viewport as it went.

Opus 5.5: isometric render on a diorama base.

What Opus built:

• Plinth: 5 m square, with blind arches along all four sides.
• Tomb: a 56.6 m square with chamfered corners. Each face has a central portal with a 5 m deep arched recess, flanked by two storeys of niches.
• Roofline: the drum and onion dome, a gilded finial reaching 73 m, four chhatris and 16 pinnacles.
• Minarets: four of them, about 43 m tall, each with three balconies.
• Materials: procedural white marble with veining, black bands for the calligraphy borders and gold for the finials.
• Print file: one watertight solid at 1:500, about 192 × 192 × 146 mm, with 1.27 million triangles.

It was not right first time. When I looked at the roof I found a hole behind each of the four portals, where the arch cut had broken through. Opus diagnosed the cause, fixed it and re-exported the STL.

Fable 5.2: the second attempt

I then switched the model to Fable 5.2 and asked it to start again from scratch in a new file: the most realistic and accurate Taj Mahal it could build, plus a printable STL.

Fable took about 25 minutes. It re-checked the published dimensions before modeling and built a noticeably more detailed scene.

Fable 5.2: isometric render, including the terrace, garden, mosque and jawab.

What Fable built:

• Proportions: a 12 m drum under a 23 m dome, a 9.6 m finial and 73 m overall, on a 95.5 m plinth.
• Recesses: all 28 arched recesses have vaulted half-dome ceilings, with doorways and lattice screens behind them.
• Roofline: the chhatris and minaret tops have open arcades, sloping eaves and lotus caps.
• Materials: marble that varies in tone slab by slab, fine dark joints on the minarets and a calligraphy-style band around each portal.
• Setting: the red sandstone terrace, the garden with its canals and central tank, the river, and simplified versions of the mosque and jawab.
• Print file: a separate build of the mausoleum with the thin parts thickened to about 1 mm, fused into one watertight shell. At 1:500 it is about 194 × 198 × 146 mm, with 1.29 million triangles.

Fable also made a mistake. In the STL I found a beam running across the top of each main arch, left there by the cornice bands. Like Opus, it found the cause and fixed both the model and the print file.

How the two compare

	Opus 5.5	Fable 5.2
Time to model and STL	About 11 minutes	About 25 minutes
Dimensions	Published figures: 95.5 m plinth, 73 m to the finial tip	Published figures, re-checked first: 12 m drum, 23 m dome, 9.6 m finial, 73 m overall
Arched recesses	Flat-backed, 5 m deep	Half-dome vaults, with doorways and lattice screens
Roofline	Solid chhatris and plain pinnacles	Open-arcade chhatris with eaves and lotus caps
Marble	One procedural marble with veining	Slab-by-slab tone variation, darker in the recesses
Setting	Small diorama base, added for the render	Terrace, garden, canals, river, mosque and jawab
Print file	Voxel-fused copy of the render model	Separate build with thin parts thickened, then voxel-fused
STL at 1:500	192 × 192 × 146 mm, 1.27 million triangles	194 × 198 × 146 mm, 1.29 million triangles
Defect I found	Holes in the roof behind all four portals	A beam across the top of all four main arches

Fable produced the better model. The vaulted recesses, the open chhatris and the slab-level marble make it read as a building, where the Opus model reads as a clean massing study. Fable also treated printing as its own problem and thickened the fragile parts, which is what someone who prints for a living would do.

Opus was more than twice as fast, and its model was correct in scale and perfectly printable. If I wanted a quick desk model, I would not need more than that.

This was not a controlled test. Fable went second, in the same session, with a brief that asked for more, so it had the benefit of the first attempt. Neither result is photoreal: the ornament is procedural, with no real calligraphy or inlay work.

What struck me most is what the two had in common. Each delivered a to-scale, watertight, printable Taj Mahal from a one-line brief in under half an hour. And each shipped one geometry defect that its own checks missed and that I caught by looking at the model. For now, a human still needs to inspect the result.

Download 3D Model: Link