CADDesigner: Conceptual CAD Model Generation with a General-Purpose Agent

Fengxiao Fan; Jingzhe Ni; Min Tang; Peng Du; Qiang Zou; Ruofeng Tong; Sirui Wang; Xiaolong Yin; Xingyu Lu

arxiv: 2508.01031 · v6 · pith:D2JAGX5Xnew · submitted 2025-08-01 · 💻 cs.AI · cs.CL

CADDesigner: Conceptual CAD Model Generation with a General-Purpose Agent

Fengxiao Fan , Jingzhe Ni , Xiaolong Yin , Sirui Wang , Xingyu Lu , Qiang Zou , Ruofeng Tong , Min Tang

show 1 more author

Peng Du

This is my paper

Pith reviewed 2026-05-21 22:59 UTC · model grok-4.3

classification 💻 cs.AI cs.CL

keywords CAD designLLM agentconceptual modelingvisual feedbackparametric modelingAI-assisted designknowledge base

0 comments

The pith

An LLM agent generates conceptual CAD models from text descriptions and sketches using iterative visual feedback.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents CADDesigner as an agent that takes textual descriptions or sketches as input and engages in dialogue to clarify requirements. It relies on a novel Explicit Context Imperative Paradigm to output CAD modeling code, then refines results through repeated visual feedback loops. The generated designs are stored in a structured knowledge base for later reuse. If successful, this setup lowers the expertise needed for early-stage parametric 3D modeling while allowing the system to accumulate design knowledge over time. Experimental comparisons show it matches or exceeds representative baselines on conceptual CAD generation tasks.

Core claim

CADDesigner, powered by large language models and built on the Explicit Context Imperative Paradigm, accepts natural language or sketch inputs, performs requirement analysis through dialogue, produces CAD code, and iteratively improves model quality via visual feedback before storing successful cases in a knowledge base.

What carries the argument

The Explicit Context Imperative Paradigm (ECIP), a prompting structure that forces the agent to maintain explicit context and issue precise imperatives when translating user input into valid CAD modeling code.

If this is right

Designers without CAD expertise can produce initial parametric models through conversation alone.
Successful designs accumulate in a knowledge base that can be queried to guide future generations.
Iterative visual feedback reduces the number of invalid or low-quality outputs compared to single-pass generation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same agent pattern could extend to other parametric modeling domains such as mechanical parts or architectural layouts.
Replacing the current vision module with stronger image-understanding models would likely cut down on feedback interpretation errors.

Load-bearing premise

The underlying LLM can correctly interpret rendered visual feedback and translate it into accurate code changes without frequent human fixes or post-processing.

What would settle it

Generate CAD models for a fixed set of 50 conceptual design prompts, render the outputs, and count how many produce valid, editable 3D models that match the original description without manual editing.

Figures

Figures reproduced from arXiv: 2508.01031 by Fengxiao Fan, Jingzhe Ni, Min Tang, Peng Du, Qiang Zou, Ruofeng Tong, Sirui Wang, Xiaolong Yin, Xingyu Lu.

**Figure 1.** Figure 1: Demonstration of various CAD models generated by CADDesigner. Our method supports multimodal input and a broad range of CAD operations, including extrusion, revolution, fillet/chamfer, sweeping, lofting, etc., as well as the creation of standard components such as flanges and screws. experimental setup and present the results. Finally, we provide a comprehensive analysis and comparison to highlight the str… view at source ↗

**Figure 2.** Figure 2: The Intelligent CAD Orchestrator Agent, CADDesigner, follows a ReAct-style paradigm to progressively transform user requirements into valid CAD models through iterative reasoning, tool execution, and feedback refinement. It first refines user requirements into detailed designs, generates executable modeling code using domain APIs, and analyzes execution results via both symbolic (e.g., shell logs and error… view at source ↗

**Figure 3.** Figure 3: Code comparison between CadQuery (left) and ECIP (right). ECIP explicitly passes context and supports standard Python constructs, improving code clarity and flexibility. 4.2. ECIP-Compliant CAD API Design ECIP (referred to as SimpleCADAPI in the actual project) is designed as a command-style Python API built on top of CadQuery’s 𝚘𝚌𝚌_𝚒𝚖𝚙𝚕.𝚜𝚑𝚊𝚙𝚎𝚜 module. It serves as an LLM-oriented intermediate representati… view at source ↗

**Figure 4.** Figure 4: Token usage and generation latency as a function of model complexity (number of commands). Models are grouped into bins of 10 commands each. 5.4.4. Effect of Model Complexity on Inference Cost We evaluate the impact of model complexity on inference cost in CADDesigner. Specifically, we focus on two main metrics: Tokens and Latency. For memory consumption, CADDesigner primarily relies on external LLM APIs, … view at source ↗

**Figure 5.** Figure 5: Average inference cost breakdown across the CADDesigner pipeline. Tokens (left) and Latency (right) for each component. used for code generation, and (2) existing text-to-CAD generation methods. 5.5.1. Comparison of CAD Representation Paradigms We first evaluate the impact of different CAD representation paradigms on code generation performance on 200 test models. Specifically, we compare our proposed EC… view at source ↗

**Figure 6.** Figure 6: Violin plots comparing the distributions of IoU, CD, and HD across ECIP, CadQuery, and build123d on 200 test models. ECIP achieves higher geometric fidelity and more consistent performance compared to the other paradigms [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗

**Figure 8.** Figure 8: Comparison of CADDesigner and cadrille on imagebased inputs. CADDesigner (blue) generates CAD models that more closely match the input images on these representative conceptual cases, benefiting from support for operations such as revolve and pattern-based constructions. In contrast, cadrille (green), relying on low-level extrusion, often produces geometrically less faithful results. structurally valid bu… view at source ↗

**Figure 7.** Figure 7: Comparison of generation results across Text2CAD, CADCodeVerify, cadrille, and our method. CADDesigner achieves the best input alignment. Text2CAD and CADCodeVerify show moderate performance, while cadrille generates syntactically valid but semantically incorrect outputs due to poor generalization from expert-level training to abstract inputs. for metric calculation. As presented in [PITH_FULL_IMAGE:figu… view at source ↗

**Figure 9.** Figure 9: Visual results with sketch-text input. Fan et al.: Preprint submitted to Elsevier Page 14 of 27 [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗

read the original abstract

Computer-Aided Design (CAD) is widely used for conceptual design and parametric 3D modeling, but typically requires a high level of expertise from designers. To lower the entry barrier and facilitate early-stage CAD modeling, we present CADDesigner, an LLM-powered agent for conceptual CAD design. The agent accepts both textual descriptions and sketches as input, engaging in interactive dialogue with users to refine and clarify design requirements through comprehensive requirement analysis. Built upon a novel Explicit Context Imperative Paradigm (ECIP), the agent generates high-quality CAD modeling code. During the generation process, the agent incorporates iterative visual feedback to improve model quality. Generated design cases can be stored in a structured knowledge base, providing a mechanism for continual knowledge accumulation and future improvement of code generation. Experimental results show that CADDesigner achieves competitive performance and outperforms representative baselines on conceptual CAD model generation tasks.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

CADDesigner packages an LLM agent with dialogue, ECIP prompting, and visual feedback for CAD code, but the performance claims rest on experiments whose objective metrics are not clear from the abstract.

read the letter

The core of this paper is a practical LLM agent for conceptual CAD that takes text or sketches, runs a dialogue to pin down requirements, generates code via the Explicit Context Imperative Paradigm, and loops in visual feedback to fix issues before storing results in a knowledge base. It reports beating some baselines on generation tasks. That combination is new enough as an integrated system for this narrow domain, even if the pieces draw from existing agent patterns. The practical goal of lowering the expertise bar for early-stage modeling is reasonable and the closed-loop visual step is a sensible engineering choice for catching geometric problems that pure text generation often misses. The knowledge-base accumulation idea also gives a path for incremental improvement without retraining. The main soft spot is the experimental section. The abstract claims competitive or better performance but supplies no numbers, no dataset description, no ablation on ECIP itself, and no concrete success rates such as fraction of code that compiles without error or quantitative similarity to reference models. If the full paper evaluates mainly through qualitative visual inspection or human preference rather than hard, reproducible metrics like valid render counts or geometric error, the outperformance claim becomes difficult to verify independently. That matches the stress-test concern and leaves the central result under-supported. The work is aimed at applied researchers and engineers building domain-specific LLM tools rather than theorists. A reader looking for concrete system designs in design automation could extract useful architecture details, but the paper will not shift broader understanding of agents or CAD. It is worth sending for peer review so referees can examine the actual evaluation protocol and any unreported quantitative results.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces CADDesigner, an LLM-powered agent for conceptual CAD design that accepts textual descriptions and sketches as input. It performs requirement analysis via interactive dialogue, generates CAD modeling code using a novel Explicit Context Imperative Paradigm (ECIP), incorporates iterative visual feedback to refine outputs, and stores successful designs in a structured knowledge base for continual improvement. The central claim is that the system achieves competitive performance and outperforms representative baselines on conceptual CAD model generation tasks.

Significance. If the experimental claims hold under rigorous quantitative scrutiny, the work could meaningfully lower barriers to early-stage CAD modeling for non-experts. The combination of visual feedback loops with a persistent knowledge base offers a practical template for LLM agents in design domains, though the absence of detailed metrics limits assessment of its advance over existing prompting and agent frameworks.

major comments (2)

[Abstract / Experimental Results] Abstract and Experimental Results section: the claim that CADDesigner 'outperforms representative baselines' is presented without any reported quantitative metrics (e.g., code compilation success rate, rendering validity rate, geometric similarity scores, or dataset size). If experiments rely primarily on qualitative visual assessment rather than objective measures of error-free CAD code generation, the performance advantage cannot be verified independently of human post-processing.
[Methodology (ECIP)] Methodology section describing ECIP: the Explicit Context Imperative Paradigm is introduced as a core contribution, yet no formal specification, pseudocode, or ablation isolating its effect versus standard iterative prompting or visual chain-of-thought is provided. Without this, it is unclear whether ECIP constitutes a load-bearing technical advance or a descriptive label for conventional agent behavior.

minor comments (2)

[System Architecture] Clarify the exact CAD language or library used (e.g., OpenSCAD, FreeCAD API) and how visual feedback is concretely implemented (screenshot analysis, error messages, or both).
[Experiments] Add explicit comparison table against baselines that reports at least success rate, iteration count, and failure modes rather than only qualitative statements.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback on our manuscript. We address each major comment below and indicate the revisions planned to strengthen the presentation of results and methodology.

read point-by-point responses

Referee: [Abstract / Experimental Results] Abstract and Experimental Results section: the claim that CADDesigner 'outperforms representative baselines' is presented without any reported quantitative metrics (e.g., code compilation success rate, rendering validity rate, geometric similarity scores, or dataset size). If experiments rely primarily on qualitative visual assessment rather than objective measures of error-free CAD code generation, the performance advantage cannot be verified independently of human post-processing.

Authors: We thank the referee for highlighting this issue. While the current manuscript emphasizes qualitative visual comparisons to illustrate the conceptual design capabilities, we agree that quantitative metrics are essential for independent verification. In the revised manuscript, we will add a table in the Experimental Results section reporting objective measures such as code compilation success rate, rendering validity rate, geometric similarity scores, and the dataset size used in evaluations. This will substantiate the claim of outperforming representative baselines with verifiable data. revision: yes
Referee: [Methodology (ECIP)] Methodology section describing ECIP: the Explicit Context Imperative Paradigm is introduced as a core contribution, yet no formal specification, pseudocode, or ablation isolating its effect versus standard iterative prompting or visual chain-of-thought is provided. Without this, it is unclear whether ECIP constitutes a load-bearing technical advance or a descriptive label for conventional agent behavior.

Authors: We appreciate the referee's point on the need for a more rigorous presentation of ECIP. To address this, the revised Methodology section will include a formal specification of the ECIP paradigm along with pseudocode detailing its key components and workflow. Furthermore, we will conduct and report an ablation study comparing ECIP against standard iterative prompting and visual chain-of-thought approaches to demonstrate its specific contributions to model quality and efficiency. revision: yes

Circularity Check

0 steps flagged

No significant circularity; engineering system evaluated externally

full rationale

The paper describes an LLM-based agent system (CADDesigner) with a novel Explicit Context Imperative Paradigm (ECIP) for generating CAD code from text and sketches, incorporating visual feedback and a knowledge base. No mathematical derivations, equations, predictions, or fitted parameters are present. Performance claims rest on experimental comparisons to external baselines rather than any self-referential definitions or self-citation chains. The evaluation is independent of the system's own outputs, satisfying the criteria for a self-contained engineering artifact with no load-bearing circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the untested premise that an LLM can correctly interpret and act on visual feedback to improve CAD code quality. No free parameters or invented physical entities are described. The ECIP paradigm is presented as a novel but unformalized modeling choice.

axioms (1)

domain assumption LLMs can generate syntactically valid CAD modeling code when guided by explicit context and visual feedback
Invoked in the description of the generation process and iterative improvement loop.

invented entities (1)

Explicit Context Imperative Paradigm (ECIP) no independent evidence
purpose: To structure the agent's context and instructions for generating CAD code
Introduced as a novel paradigm in the abstract; no independent evidence or formal definition provided beyond the name.

pith-pipeline@v0.9.0 · 5695 in / 1337 out tokens · 37055 ms · 2026-05-21T22:59:53.893857+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Built upon a novel Explicit Context Imperative Paradigm (ECIP), the agent generates high-quality CAD modeling code... iterative visual feedback
IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

ReAct-style agent... Requirement Refinement Tool, Code Generation Tool, Visual Feedback Tool

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Agent-Aided Design for Dynamic CAD Models
cs.AI 2026-04 unverdicted novelty 6.0

AADvark extends agent-aided CAD design to dynamic 3D assemblies with movable parts by integrating constraint solvers and visual feedback to create a verification signal for the agent.
Memory-Augmented Reinforcement Learning Agent for CAD Generation
cs.AI 2026-05 unverdicted novelty 5.0

Memory-augmented RL agent with case and skill libraries plus dynamic retrieval improves success rate and geometric consistency for complex CAD model generation.
Self-Improving CAD Generation Agents with Finite Element Analysis as Feedback
cs.GR 2026-05 unverdicted novelty 5.0

CAD agents using finite element analysis feedback plus new text blueprint and multi-view image signals improve geometric accuracy on S2O and Fusion360 benchmarks while addressing physical validity gaps in prior genera...

Reference graph

Works this paper leans on

28 extracted references · 28 canonical work pages · cited by 3 Pith papers · 1 internal anchor

[1]

DeepCAD: A deep generative network for computer-aided design models

Rundi Wu, Chang Xiao, and Changxi Zheng. DeepCAD: A deep generative network for computer-aided design models. InProceedings oftheIEEE/CVFInternationalConferenceonComputerVision ,pages 6772–6782, 2021

work page 2021
[2]

Mamba- CAD: State space model for 3D computer-aided design generative modeling

Xueyang Li, Yunzhong Lou, Yu Song, and Xiangdong Zhou. Mamba- CAD: State space model for 3D computer-aided design generative modeling. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 5013–5021, 2025

work page 2025
[3]

CAD-SIGNet: CAD language inference from point clouds using layer-wise sketch instance guided attention

Mohammad Sadil Khan, Elona Dupont, Sk Aziz Ali, Kseniya Cherenkova, Anis Kacem, and Djamila Aouada. CAD-SIGNet: CAD language inference from point clouds using layer-wise sketch instance guided attention. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4713–4722, 2024

work page 2024
[4]

Text2CAD: Generating sequential CAD designs from beginner-to-expert level text prompts

Mohammad Sadil Khan, Sankalp Sinha, Talha Uddin Sheikh, Didier Stricker, Sk Aziz Ali, and Muhammad Zeshan Afzal. Text2CAD: Generating sequential CAD designs from beginner-to-expert level text prompts. In Advances in Neural Information Processing Systems, volume 37, pages 7552–7579, 2024

work page 2024
[5]

CAD-Llama: leveraging large language models for computer-aided design parametric 3D model generation

Jiahao Li, Weijian Ma, Xueyang Li, Yunzhong Lou, Guichun Zhou, and Xiangdong Zhou. CAD-Llama: leveraging large language models for computer-aided design parametric 3D model generation. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 18563–18573, 2025

work page 2025
[6]

CAD-GPT: Synthesising CAD construction sequence with spatial reasoning-enhanced multimodal LLMs

Siyu Wang, Cailian Chen, Xinyi Le, Qimin Xu, Lei Xu, Yanzhou Zhang, and Jie Yang. CAD-GPT: Synthesising CAD construction sequence with spatial reasoning-enhanced multimodal LLMs. InPro- ceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 7880–7888, 2025

work page 2025
[7]

CAD-Recode: Reverse engineering CAD code from point clouds.arXiv preprint arXiv:2412.14042, 2024

Danila Rukhovich, Elona Dupont, Dimitrios Mallis, Kseniya Cherenkova, Anis Kacem, and Djamila Aouada. CAD-Recode: Reverse engineering CAD code from point clouds.arXiv preprint arXiv:2412.14042, 2024

work page arXiv 2024
[8]

CadQuery, February 2026

CadQuery contributors. CadQuery, February 2026

work page 2026
[9]

Karl D. D. Willis, Yewen Pu, Jieliang Luo, Hang Chu, Tao Du, JosephG.Lambourne,ArmandoSolar-Lezama,andWojciechMatusik. Fusion360gallery:AdatasetandenvironmentforprogrammaticCAD construction from human design sequences.ACM Transactions on Graphics (TOG), 40(4), 2021

work page 2021
[10]

SkexGen:Autore- gressive generation of CAD construction sequences with disentangled codebooks

Xiang Xu, Karl DD Willis, Joseph G Lambourne, Chin-Yi Cheng, PradeepKumarJayaraman,andYasutakaFurukawa. SkexGen:Autore- gressive generation of CAD construction sequences with disentangled codebooks. InInternational Conference on Machine Learning, pages 24698–24724, 2022

work page 2022
[11]

Diffusion-CAD: Controllable diffusion model for generating computer-aided design models.IEEE Transactions on Visualization and Computer Graphics, 2025

Aijia Zhang, Weiqiang Jia, Qiang Zou, Yixiong Feng, Xiaoxiang Wei, and Ye Zhang. Diffusion-CAD: Controllable diffusion model for generating computer-aided design models.IEEE Transactions on Visualization and Computer Graphics, 2025

work page 2025
[12]

ComplexGen: CAD reconstruction by B-rep chain complex generation

HaoxiangGuo, Shilin Liu, Hao Pan, Yang Liu, Xin Tong, and Baining Guo. ComplexGen: CAD reconstruction by B-rep chain complex generation. ACM Transactions on Graphics (TOG), 41(4):1–18, 2022

work page 2022
[13]

Draw step by step: reconstructing CAD construction sequences from point clouds via multimodal diffusion

Weijian Ma, Shuaiqi Chen, Yunzhong Lou, Xueyang Li, and Xiang- dong Zhou. Draw step by step: reconstructing CAD construction sequences from point clouds via multimodal diffusion. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 27154–27163, 2024

work page 2024
[14]

CADDreamer: CAD object generation from single-view images

YuanLi,ChengLin,YuanLiu,XiaoxiaoLong,ChenxuZhang,Ningna Wang, Xin Li, Wenping Wang, and Xiaohu Guo. CADDreamer: CAD object generation from single-view images. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 21448– 21457, 2025

work page 2025
[15]

CADCrafter: Generating computer- aided design models from unconstrained images

ChengChen,JiachengWei,TianrunChen,ChiZhang,XiaofengYang, Shangzhan Zhang, Bingchen Yang, Chuan-Sheng Foo, Guosheng Lin, Qixing Huang, and Fayao Liu. CADCrafter: Generating computer- aided design models from unconstrained images. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 11073–11082, 2025

work page 2025
[16]

CAD-LLM: Large language model for cad generation

Sifan Wu, Amir Khasahmadi, Mor Katz, Pradeep Kumar Jayaraman, Yewen Pu, Karl Willis, and Bang Liu. CAD-LLM: Large language model for cad generation. InProceedings of the neural information processing systems conference, 2023

work page 2023
[17]

CadVLM: Bridging Fan et al.:Preprint submitted to Elsevier Page 15 of 16 language and vision in the generation of parametric CAD sketches

Sifan Wu, Amir Hosein Khasahmadi, Mor Katz, Pradeep Kumar Jayaraman, Yewen Pu, Karl Willis, and Bang Liu. CadVLM: Bridging Fan et al.:Preprint submitted to Elsevier Page 15 of 16 language and vision in the generation of parametric CAD sketches. In European Conference on Computer Vision, pages 368–384, 2024

work page 2024
[18]

Cad-mllm: Unifying multimodality-conditioned cad generation with mllm

JingweiXu,ZiboZhao,ChenyuWang,WenLiu,YiMa,andShenghua Gao. CAD-MLLM: Unifying multimodality-conditioned CAD gener- ation with MLLM.arXiv preprint arXiv:2411.04954, 2024

work page arXiv 2024
[19]

Qwen2 Technical Report

Qwen Team. Qwen2 technical report. arXiv preprint arXiv:2407.10671, 2024

work page internal anchor Pith review Pith/arXiv arXiv 2024
[20]

Text-to-CadQuery: A New Paradigm for CADgenerationwithscalablelargemodelcapabilities

Haoyang Xie and Feng Ju. Text-to-CadQuery: A New Paradigm for CADgenerationwithscalablelargemodelcapabilities. arXivpreprint arXiv:2505.06507, 2025

work page arXiv 2025
[21]

CAD-Assistant: Tool-augmented VLLMs as generic CAD task solvers.arXiv preprint arXiv:2412.13810, 2024

Dimitrios Mallis, Ahmet Serdar Karadeniz, Sebastian Cavada, Danila Rukhovich, Niki Foteinopoulou, Kseniya Cherenkova, Anis Kacem, and Djamila Aouada. CAD-Assistant: Tool-augmented VLLMs as generic CAD task solvers.arXiv preprint arXiv:2412.13810, 2024

work page arXiv 2024
[22]

FreeCAD:Open- source parametric 3D CAD modeler, 2024

JuergenRiegel,WernerMayer,andYorikvanHavre. FreeCAD:Open- source parametric 3D CAD modeler, 2024

work page 2024
[23]

Seek-CAD:Aself-refinedgenerativemodelingfor3DparametricCAD using local inference via DeepSeek.arXiv preprint arXiv:2505.17702, 2025

XueyangLi,JiahaoLi,YuSong,YunzhongLou,andXiangdongZhou. Seek-CAD:Aself-refinedgenerativemodelingfor3DparametricCAD using local inference via DeepSeek.arXiv preprint arXiv:2505.17702, 2025

work page arXiv 2025
[24]

Hao Zhang, Feng Li, Shilong Liu, Lei Zhang, Hang Su, Jun Zhu, Lionel M

ZeqingYuan,HaoxuanLan,QiangZou,andJunboZhao. 3D-PreMise: Can large language models generate 3D shapes with sharp features and parametric control?arXiv preprint arXiv:2401.06437, 2024

work page arXiv 2024
[25]

Generating CAD code with vision-language models for 3d designs.arXiv preprint arXiv:2410.05340, 2024

Kamel Alrashedy, Pradyumna Tambwekar, Zulfiqar Zaidi, Megan Langwasser, Wei Xu, and Matthew Gombolay. Generating CAD code with vision-language models for 3d designs.arXiv preprint arXiv:2410.05340, 2024

work page arXiv 2024
[26]

Roger Maitland, jdegenstein, Bernhard, Ethan Rooke, JR Mobley, snoyer, Jojain, Andreas Felix Häberle, Ruud, Ami Fischman, Ja- son S. McMullan, Roman Dvořák, simon klemenc, BogdanTheGeek, Spectre5, Dalibor Frívaldský, Daniele D’Orazio, George, hoijui, OpenVMP, Yeicor, Alexander Steppke, mayhem 64, luzpaz, nobkd, Victor Poughon, slobberingant, Arno Bosch, B...

work page 2025
[27]

ReAct: Synergizing reasoning and acting in language models

ShunyuYao,JeffreyZhao,DianYu,NanDu,IzhakShafran,KarthikR Narasimhan, and Yuan Cao. ReAct: Synergizing reasoning and acting in language models. InThe Eleventh International Conference on Learning Representations, 2023

work page 2023
[28]

cadrille: Multi-modal CADreconstructionwithonlinereinforcementlearning

Maksim Kolodiazhnyi, Denis Tarasov, Dmitrii Zhemchuzhnikov, Alexander Nikulin, Ilya Zisman, Anna Vorontsova, Anton Konushin, Vladislav Kurenkov, and Danila Rukhovich. cadrille: Multi-modal CADreconstructionwithonlinereinforcementlearning. arXivpreprint arXiv:2505.22914, 2025. Fan et al.:Preprint submitted to Elsevier Page 16 of 16

work page arXiv 2025

[1] [1]

DeepCAD: A deep generative network for computer-aided design models

Rundi Wu, Chang Xiao, and Changxi Zheng. DeepCAD: A deep generative network for computer-aided design models. InProceedings oftheIEEE/CVFInternationalConferenceonComputerVision ,pages 6772–6782, 2021

work page 2021

[2] [2]

Mamba- CAD: State space model for 3D computer-aided design generative modeling

Xueyang Li, Yunzhong Lou, Yu Song, and Xiangdong Zhou. Mamba- CAD: State space model for 3D computer-aided design generative modeling. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 5013–5021, 2025

work page 2025

[3] [3]

CAD-SIGNet: CAD language inference from point clouds using layer-wise sketch instance guided attention

Mohammad Sadil Khan, Elona Dupont, Sk Aziz Ali, Kseniya Cherenkova, Anis Kacem, and Djamila Aouada. CAD-SIGNet: CAD language inference from point clouds using layer-wise sketch instance guided attention. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4713–4722, 2024

work page 2024

[4] [4]

Text2CAD: Generating sequential CAD designs from beginner-to-expert level text prompts

Mohammad Sadil Khan, Sankalp Sinha, Talha Uddin Sheikh, Didier Stricker, Sk Aziz Ali, and Muhammad Zeshan Afzal. Text2CAD: Generating sequential CAD designs from beginner-to-expert level text prompts. In Advances in Neural Information Processing Systems, volume 37, pages 7552–7579, 2024

work page 2024

[5] [5]

CAD-Llama: leveraging large language models for computer-aided design parametric 3D model generation

Jiahao Li, Weijian Ma, Xueyang Li, Yunzhong Lou, Guichun Zhou, and Xiangdong Zhou. CAD-Llama: leveraging large language models for computer-aided design parametric 3D model generation. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 18563–18573, 2025

work page 2025

[6] [6]

CAD-GPT: Synthesising CAD construction sequence with spatial reasoning-enhanced multimodal LLMs

Siyu Wang, Cailian Chen, Xinyi Le, Qimin Xu, Lei Xu, Yanzhou Zhang, and Jie Yang. CAD-GPT: Synthesising CAD construction sequence with spatial reasoning-enhanced multimodal LLMs. InPro- ceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 7880–7888, 2025

work page 2025

[7] [7]

CAD-Recode: Reverse engineering CAD code from point clouds.arXiv preprint arXiv:2412.14042, 2024

Danila Rukhovich, Elona Dupont, Dimitrios Mallis, Kseniya Cherenkova, Anis Kacem, and Djamila Aouada. CAD-Recode: Reverse engineering CAD code from point clouds.arXiv preprint arXiv:2412.14042, 2024

work page arXiv 2024

[8] [8]

CadQuery, February 2026

CadQuery contributors. CadQuery, February 2026

work page 2026

[9] [9]

Karl D. D. Willis, Yewen Pu, Jieliang Luo, Hang Chu, Tao Du, JosephG.Lambourne,ArmandoSolar-Lezama,andWojciechMatusik. Fusion360gallery:AdatasetandenvironmentforprogrammaticCAD construction from human design sequences.ACM Transactions on Graphics (TOG), 40(4), 2021

work page 2021

[10] [10]

SkexGen:Autore- gressive generation of CAD construction sequences with disentangled codebooks

Xiang Xu, Karl DD Willis, Joseph G Lambourne, Chin-Yi Cheng, PradeepKumarJayaraman,andYasutakaFurukawa. SkexGen:Autore- gressive generation of CAD construction sequences with disentangled codebooks. InInternational Conference on Machine Learning, pages 24698–24724, 2022

work page 2022

[11] [11]

Diffusion-CAD: Controllable diffusion model for generating computer-aided design models.IEEE Transactions on Visualization and Computer Graphics, 2025

Aijia Zhang, Weiqiang Jia, Qiang Zou, Yixiong Feng, Xiaoxiang Wei, and Ye Zhang. Diffusion-CAD: Controllable diffusion model for generating computer-aided design models.IEEE Transactions on Visualization and Computer Graphics, 2025

work page 2025

[12] [12]

ComplexGen: CAD reconstruction by B-rep chain complex generation

HaoxiangGuo, Shilin Liu, Hao Pan, Yang Liu, Xin Tong, and Baining Guo. ComplexGen: CAD reconstruction by B-rep chain complex generation. ACM Transactions on Graphics (TOG), 41(4):1–18, 2022

work page 2022

[13] [13]

Draw step by step: reconstructing CAD construction sequences from point clouds via multimodal diffusion

Weijian Ma, Shuaiqi Chen, Yunzhong Lou, Xueyang Li, and Xiang- dong Zhou. Draw step by step: reconstructing CAD construction sequences from point clouds via multimodal diffusion. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 27154–27163, 2024

work page 2024

[14] [14]

CADDreamer: CAD object generation from single-view images

YuanLi,ChengLin,YuanLiu,XiaoxiaoLong,ChenxuZhang,Ningna Wang, Xin Li, Wenping Wang, and Xiaohu Guo. CADDreamer: CAD object generation from single-view images. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 21448– 21457, 2025

work page 2025

[15] [15]

CADCrafter: Generating computer- aided design models from unconstrained images

ChengChen,JiachengWei,TianrunChen,ChiZhang,XiaofengYang, Shangzhan Zhang, Bingchen Yang, Chuan-Sheng Foo, Guosheng Lin, Qixing Huang, and Fayao Liu. CADCrafter: Generating computer- aided design models from unconstrained images. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 11073–11082, 2025

work page 2025

[16] [16]

CAD-LLM: Large language model for cad generation

Sifan Wu, Amir Khasahmadi, Mor Katz, Pradeep Kumar Jayaraman, Yewen Pu, Karl Willis, and Bang Liu. CAD-LLM: Large language model for cad generation. InProceedings of the neural information processing systems conference, 2023

work page 2023

[17] [17]

CadVLM: Bridging Fan et al.:Preprint submitted to Elsevier Page 15 of 16 language and vision in the generation of parametric CAD sketches

Sifan Wu, Amir Hosein Khasahmadi, Mor Katz, Pradeep Kumar Jayaraman, Yewen Pu, Karl Willis, and Bang Liu. CadVLM: Bridging Fan et al.:Preprint submitted to Elsevier Page 15 of 16 language and vision in the generation of parametric CAD sketches. In European Conference on Computer Vision, pages 368–384, 2024

work page 2024

[18] [18]

Cad-mllm: Unifying multimodality-conditioned cad generation with mllm

JingweiXu,ZiboZhao,ChenyuWang,WenLiu,YiMa,andShenghua Gao. CAD-MLLM: Unifying multimodality-conditioned CAD gener- ation with MLLM.arXiv preprint arXiv:2411.04954, 2024

work page arXiv 2024

[19] [19]

Qwen2 Technical Report

Qwen Team. Qwen2 technical report. arXiv preprint arXiv:2407.10671, 2024

work page internal anchor Pith review Pith/arXiv arXiv 2024

[20] [20]

Text-to-CadQuery: A New Paradigm for CADgenerationwithscalablelargemodelcapabilities

Haoyang Xie and Feng Ju. Text-to-CadQuery: A New Paradigm for CADgenerationwithscalablelargemodelcapabilities. arXivpreprint arXiv:2505.06507, 2025

work page arXiv 2025

[21] [21]

CAD-Assistant: Tool-augmented VLLMs as generic CAD task solvers.arXiv preprint arXiv:2412.13810, 2024

Dimitrios Mallis, Ahmet Serdar Karadeniz, Sebastian Cavada, Danila Rukhovich, Niki Foteinopoulou, Kseniya Cherenkova, Anis Kacem, and Djamila Aouada. CAD-Assistant: Tool-augmented VLLMs as generic CAD task solvers.arXiv preprint arXiv:2412.13810, 2024

work page arXiv 2024

[22] [22]

FreeCAD:Open- source parametric 3D CAD modeler, 2024

JuergenRiegel,WernerMayer,andYorikvanHavre. FreeCAD:Open- source parametric 3D CAD modeler, 2024

work page 2024

[23] [23]

Seek-CAD:Aself-refinedgenerativemodelingfor3DparametricCAD using local inference via DeepSeek.arXiv preprint arXiv:2505.17702, 2025

XueyangLi,JiahaoLi,YuSong,YunzhongLou,andXiangdongZhou. Seek-CAD:Aself-refinedgenerativemodelingfor3DparametricCAD using local inference via DeepSeek.arXiv preprint arXiv:2505.17702, 2025

work page arXiv 2025

[24] [24]

Hao Zhang, Feng Li, Shilong Liu, Lei Zhang, Hang Su, Jun Zhu, Lionel M

ZeqingYuan,HaoxuanLan,QiangZou,andJunboZhao. 3D-PreMise: Can large language models generate 3D shapes with sharp features and parametric control?arXiv preprint arXiv:2401.06437, 2024

work page arXiv 2024

[25] [25]

Generating CAD code with vision-language models for 3d designs.arXiv preprint arXiv:2410.05340, 2024

Kamel Alrashedy, Pradyumna Tambwekar, Zulfiqar Zaidi, Megan Langwasser, Wei Xu, and Matthew Gombolay. Generating CAD code with vision-language models for 3d designs.arXiv preprint arXiv:2410.05340, 2024

work page arXiv 2024

[26] [26]

Roger Maitland, jdegenstein, Bernhard, Ethan Rooke, JR Mobley, snoyer, Jojain, Andreas Felix Häberle, Ruud, Ami Fischman, Ja- son S. McMullan, Roman Dvořák, simon klemenc, BogdanTheGeek, Spectre5, Dalibor Frívaldský, Daniele D’Orazio, George, hoijui, OpenVMP, Yeicor, Alexander Steppke, mayhem 64, luzpaz, nobkd, Victor Poughon, slobberingant, Arno Bosch, B...

work page 2025

[27] [27]

ReAct: Synergizing reasoning and acting in language models

ShunyuYao,JeffreyZhao,DianYu,NanDu,IzhakShafran,KarthikR Narasimhan, and Yuan Cao. ReAct: Synergizing reasoning and acting in language models. InThe Eleventh International Conference on Learning Representations, 2023

work page 2023

[28] [28]

cadrille: Multi-modal CADreconstructionwithonlinereinforcementlearning

Maksim Kolodiazhnyi, Denis Tarasov, Dmitrii Zhemchuzhnikov, Alexander Nikulin, Ilya Zisman, Anna Vorontsova, Anton Konushin, Vladislav Kurenkov, and Danila Rukhovich. cadrille: Multi-modal CADreconstructionwithonlinereinforcementlearning. arXivpreprint arXiv:2505.22914, 2025. Fan et al.:Preprint submitted to Elsevier Page 16 of 16

work page arXiv 2025