Building the future of 3D generation.
-1
Joined December 2022
- Tweets89.1K
- Following2K
- Followers2.4K
- Likes37.7K
C33D retweeted
@c33dbench is an Artificial Parametric Design Project where LLM is prompted to design 3D objects, shapes, structures, geometries - and the output is a graph based, non-destructive, editable 3D model.
Heres the most interesting output yet by Kimi 2.7 @kimi_moonshot @kimidevs when asked to build own memory palace and it was way ahead of any other open source frontier model - only glm 5.2 made a working graph visible in 3D - other models such as nemotron super, minimax m3, qwen 3.5, deepseek v4 flash - failed - ran via ollama at the local run of C33D bench.
Cc @Blender @AutodeskRevit @dynamobim @McNeelEurope @RhinoGrass @nvidiaai
C33D retweeted
Launched on @RobinhoodCrypto w @bankrbot
0x412bf6ABA5C55533C89C15712713Bd90e2aDabA3
🧵 C33D — Evaluating Parametric Design Intelligence in AI
Evaluating AI spatial reasoning has a major blind spot: existing benchmarks focus entirely on static geometry or code compilation. They do not evaluate if the model's design survives changes.
We are launching C33D, an open-science platform and benchmark for AI-native parametric CAD.
🔗 github.com/3esign/c33d
C33D retweeted
Yo @robinhoodapp u need more stuff like this
Launched C33D on @RobinHood -
to support development of unique open-source parametric geometry and graph intelligence harness.
@c33dbench
github.com/3esign/c33d c33d.vercel.app
C33D retweeted
C33D on RobinHood
0x412bf6ABA5C55533C89C15712713Bd90e2aDabA3
@bankrbot launch C33D on RobinHood - name C33D AI Benchmark - to support development of unique open-source parametric geometry and graph intelligence harness - include github.com/3esign/c33d and c33d.vercel.app and x @c33dbench
C33D retweeted
Launched C33D on @RobinhoodCrypto chain
0x412bf6ABA5C55533C89C15712713Bd90e2aDabA3
github.com/3esign/c33d
to support the development of a browser-based, AI-native parametric CAD system. You describe an object in chat, and an LLM agent architects it as a node graph. Because the output is a graph and not a static mesh, the model remains fully parametric — you can tweak dimensions, drag sliders, and remix features without breaking the design.
In the RobinHood benchmark (make a robin hood) best result came from @Zai_org GLM 5.2. In the comments you can see how other models perform.
The underlying engine runs OpenCascade (via replicad) in WebAssembly — producing robust B-Rep solids exportable directly to STEP for manufacturing or 3D printing.
C33D retweeted
RobinHood by GLM 5.2 @zai_org on C33D 0x412bf6ABA5C55533C89C15712713Bd90e2aDabA3
🧵 C33D — Evaluating Parametric Design Intelligence in AI
Evaluating AI spatial reasoning has a major blind spot: existing benchmarks focus entirely on static geometry or code compilation. They do not evaluate if the model's design survives changes.
We are launching C33D, an open-science platform and benchmark for AI-native parametric CAD.
🔗 github.com/3esign/c33d
C33D retweeted
C33D @c33dbench is the only AI/CAD project which is also opensource and enables creation of non destructible 3D geometry by designing graphs. And as a support we launched a token on @RobinhoodCrypto to help us advance it further. Cc @vladtenev
Launched C33D on @RobinHood -
to support development of unique open-source parametric geometry and graph intelligence harness.
@c33dbench
github.com/3esign/c33d c33d.vercel.app