mlsanityData to 14 Oct 2026

3DHarnessBench: Probing Agentic 3D-to-Code Capabilities of Frontier Vision-Language Models

arXiv 2609.06535 ↗First posted 6 Sep 2026cs.CVcited by 1

On the map: Multimodal LLMs / vision-language models → Embodied and 3D vision-language models

also in 1 more
Abstract (arXiv)

We introduce 3DHarnessBench, a benchmark that evaluates the agentic ability of frontier vision-language models (VLMs) to recover 3D geometry as Blender Python code from a variety of inputs. Unlike previous frameworks that prompt the VLMs with a fixed input (e.g., a single rendering or a text description), 3DHarnessBench evaluates four separate harness settings that progressively enable active agentic exploration, facilitated by recent Blender MCP functionality. Our hierarchy from Single-view, Multi-view, Active Visual (arbitrary viewpoint access), and Full 3D Interaction (complete access to the target object through Blender function calls) probes the models' abilities in both visual perception and active inference, tool calling, and self-correction. We observe that the ability of all frontier models to recover 3D geometry improves significantly with richer function call access, although the improvements are strongly model-dependent, revealing highly uneven agentic 3D-to-code capabilities. We will release the benchmark, code, outputs, and agent trajectories for reproducible 3D evaluation.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Embodied and 3D vision-language models, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 55 of its 71 references that are in the map. First posted on arXiv 6 Sep 2026.

Builds on

55 of 71 references.

Closest ancestors · 13

Same direction, newest first

  1. 2026SceneActBench: Can Agents Act on the 3D Scenes They See?24 Jul
  2. 2026P3D-Bench: Benchmarking MLLMs for Parametric 3D Generation and Structural Reasoning9 Jun
  3. 2026ReVSI: Rebuilding Visual Spatial Intelligence Evaluation for Accurate Assessment of VLM 3D Reasoning27 Apr
  4. 2026Spa3R: Predictive Spatial Field Modeling for 3D Visual Reasoning24 Feb
  5. 2025N3D-VLM: Native 3D Grounding Enables Accurate Spatial Reasoning in Vision-Language Models18 Dec
  6. 2025G2VLM: Geometry Grounded Vision Language Model with Unified 3D Reconstruction and Spatial Reasoning26 Nov
  7. 2025MMSI-Bench: A Benchmark for Multi-Image Spatial Intelligence29 May
  8. 2025ViewSpatial-Bench: Evaluating Multi-perspective Spatial Localization in Vision-Language Models27 May
  9. 2025VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction26 May
  10. 2025SpatialLLM: A Compound 3D-Informed Design towards Spatially-Intelligent Large Multimodal Models1 May
  11. 2025MM-Spatial: Exploring 3D Spatial Understanding in Multimodal LLMs17 Mar
  12. 2024Thinking in Space: How Multimodal Large Language Models See, Remember, and Recall Spaces18 Dec
1 more
  1. 2024SpatialVLM: Endowing Vision-Language Models with Spatial Reasoning Capabilities22 Jan

Other references · 42

  1. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,974
  2. 2025DINOv3cited by 1,414
  3. 2025SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Featurescited by 1,188
  4. 2016A Point Set Generation Network for 3D Object Reconstruction from a Single Imagecited by 979
  5. 2023LRM: Large Reconstruction Model for Single Image to 3Dcited by 849
  6. 2024Structured 3D Latents for Scalable and Versatile 3D Generationcited by 770
  7. 2024InstantMesh: Efficient 3D Mesh Generation from a Single Image with Sparse-view Large Reconstruction Modelscited by 433
  8. 2023PointLLM: Empowering Large Language Models to Understand Point Cloudscited by 333
  9. 2025SAM 3D: 3Dfy Anything in Imagescited by 247
  10. 2023Uni3D: Exploring Unified 3D Representation at Scalecited by 237
  11. 2024TripoSR: Fast 3D Object Reconstruction from a Single Imagecited by 232
  12. 2025TripoSG: High-Fidelity 3D Shape Synthesis using Large-Scale Rectified Flow Modelscited by 213
30 more
  1. 2019Topology-Preserving Deep Image Segmentationcited by 198
  2. 2024GPT-4V(ision) is a Human-Aligned Evaluator for Text-to-3D Generationcited by 153
  3. 20243DSRBench: A Comprehensive 3D Spatial Reasoning Benchmarkcited by 139
  4. 2024SceneCraft: An LLM Agent for Synthesizing 3D Scene as Blender Codecited by 134
  5. 20233D-GPT: Procedural 3D Modeling with Large Language Modelscited by 95
  6. 2026Vision-as-Inverse-Graphics Agent via Interleaved Multimodal Reasoningcited by 49
  7. 2024Img2CAD: Conditioned 3D CAD Model Generation from Single Image with Structured Visual Geometrycited by 37
  8. 2025BlenderGym: Benchmarking Foundational Model Systems for Graphics Editingcited by 31
  9. 2024BlenderLLM: Training Large Language Models for Computer-Aided Design with Self-improvementcited by 29
  10. 2024NARUTO: Neural Active Reconstruction from Uncertain Target Observationscited by 28
  11. 2024Re-Thinking Inverse Graphics With Large Language Modelscited by 23
  12. 2025LL3M: Large Language 3D Modelerscited by 20
  13. 2025MeshCoder: LLM-Powered Structured Mesh Code Generation from Point Cloudscited by 19
  14. 20233DAxiesPrompts: Unleashing the 3D Spatial Task Capabilities of GPT-4Vcited by 15
  15. 2026BenchCAD: A Comprehensive, Industry-Standard Benchmark for Programmatic CADcited by 15
  16. 20263DCodeBench: Benchmarking Agentic Procedural 3D Modeling Via Codecited by 15
  17. 2026Text2CAD-Bench: A Benchmark for LLM-based Text-to-Parametric CAD Generationcited by 12
  18. 2026CADBench: A Multimodal Benchmark for AI-Assisted CAD Program Generationcited by 11
  19. 2026Text-to-CAD Evaluation with CADTestscited by 10
  20. 2026Thinking in Blender: Staged Executable Inverse Graphics with Vision-Language Modelscited by 10
  21. 2026VoxelCodeBench: Benchmarking 3D World Modeling Through Code Generationcited by 8
  22. 2026MUSE: Benchmarking Manufacturable, Functional, and Assemblable Text-to-CAD Generationcited by 8
  23. 2026WorldClaw: Agentic 3D Open-World Generation at Scalecited by 8
  24. 2025AREA3D: Active Reconstruction Agent with Unified Feed-Forward 3D Perception and Vision-Language Guidancecited by 4
  25. 2026EZBlender: Efficient 3D Editing with Plan-and-ReAct Agentcited by 4
  26. 2026MultiView-Bench: A Diagnostic Benchmark for World-Centric Multi-View Integration in VLMscited by 4
  27. 20253Dify: a Framework for Procedural 3D-CG Generation Assisted by LLMs Using MCP and RAGcited by 3
  28. 2026UniCAD: A Unified Benchmark and Universal Model for Multi-Modal Multi-Task CADcited by 3
  29. 2026Nova3D: Code-Native Generation of Programmable 3D Assetscited by 2
  30. 20263D-CoS: A New 3D Reconstruction Paradigm Based on VLM Code Synthesiscited by 1

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026CineRecon: Benchmarking Frontier Models for Agentic Cinematic Reconstructioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025IR3D-Bench: Evaluating Vision-Language Model Scene Understanding as Agentic Inverse Rendering
  2. 2026ViSculpt: Visual-Centric Agentic Geometry Editing
  3. 2024Agent3D-Zero: An Agent for Zero-shot 3D Understanding
  4. 2025Abstract 3D Perception for Spatial Intelligence in Vision-Language Models
  5. 2025VULCAN: Tool-Augmented Multi Agents for Iterative 3D Object Arrangement
  6. 2025Let Language Constrain Geometry: Vision-Language Models as Semantic and Spatial Critics for 3D Generation
  7. 20253D-Aware Vision-Language Models Fine-Tuning with Geometric Distillation
  8. 2026Do Vision-Language Models Understand 3D Scenes or Just Catalogue Objects?