Positioning
Hunyuan3D-Buffalo 1.0 is Tencent Hunyuan3D’s unified multimodal research framework for 3D generation, understanding and editing. It combines a Hunyuan3D-VLM backbone with Hunyuan3D diffusion transformer modules. (tencent-hunyuan.github.io)
Capabilities and uses
- 3D understanding: Answers questions about 3D objects and supports language-guided grounding and spatial reasoning.
- Text-to-3D: Generates 3D assets from text descriptions.
- Instruction-guided editing: Modifies an existing 3D object according to natural-language instructions, with conditioning intended to preserve its overall structure and unedited regions.
- Part generation: Uses language to identify, extract or generate semantic parts of a 3D object.
These capabilities can support 3D asset creation, inspection, editing and decomposition workflows. The research paper reports training on an 87-million-sample multimodal corpus. (tencent-hunyuan.github.io)
Limitations and availability
The paper says the current model focuses primarily on geometry editing; end-to-end texture editing remains largely unexplored. It also notes that ambiguities in training descriptions can introduce noise and that consistency within an edited region is not guaranteed. The official project page labels code as “Stay Tuned”; the cited announcement does not establish a public API, model weights, deployment requirements or a model-specific API identifier. (arxiv.org)
Release date
Tencent’s repository announces the paper and project page on August 5, 2026. The paper’s initial arXiv submission is dated August 3, 2026; these are different publication events. (github.com)
Pricing
As checked on November 16, 2026, model-specific input and output billing for Hunyuan3D-Buffalo 1.0 could not be verified in either CNY or USD. Tencent’s mainland China Hunyuan 3D pricing page and international Hunyuan 3D pricing page describe separate 3D services, but neither identifies Buffalo 1.0. Their rates therefore cannot be attributed to this model. (cloud.tencent.com)