AITOPAI热报
Models

Hi3D V1.0 开源：训练资产扩至 200 万，切片成功率达 90%

kimmonismus@kimmonismus
Post
View the post
Why we picked it

做 3D 资产的朋友可以试试，Hi3D V1.0 刚开源，训练数据翻了四倍，擅长补全正面视角看不到的细节。

Hi3D（前身为 Sparc3D）发布并开源了 V1.0 模型的代码和权重。该模型的训练资产从去年的 50 万增加到约 200 万个。团队称 V1.0 已在实际生产流程中得到验证，技术报告显示其在多视角保真度和细节丰富度上领先对比模型，切片成功率为 90%。模型主打保留正面渲染容易隐藏的小部件和不可见侧面，目前已提供免费在线体验。

Original · kimmonismus

Hi3D V1.0 (formerly Sparc3D) is building on last year's 3D research with roughly 2 million training assets, up from 500,000, according to the team.

Hi3D (@Hi3D_AI) has now released the code and weights of V1.0, a model the team says has already been validated in real production workflows.

In the team's technical report, V1.0 leads the compared models on multi-view fidelity and detail richness, with a 90% slicing success rate.

For anyone building 3D assets, the promise here is preserving the small parts and unseen sides that a front-view render can hide.

Release (try it free online now, no deployment needed): https://t.co/vr87estMqt GitHub: https://t.co/Re1Z6EUOEa

#Hi3D#Sparc3D#开源模型#3D生成#多模态

Keep reading

Recent stories with the same tag or section
ModelsUS vs ChinaSelected21:54
NVIDIA Nemotron 3.5 Super VL Scores 1.94% on ARC-AGI-2

The pricing is the notable part: at two to three cents per task, Nemotron 3.5 Super VL is positioned as a low-cost vision-language option even though its abstract-reasoning scores trail the leaders. Teams benchmarking multimodal models against ARC-AGI can compare cost-performance trade-offs directly, since ARC Prize reports price alongside accuracy for every entry.

#NVIDIA#Nemotron 3.5 Super VL#ARC-AGI#reasoning models
官方账号ARC Prize@arcprizeSource
ModelsBuildersSelected21:30
Sakana AI Outlines Three-Pillar Strategy to Rebuild Model Layer

Sakana's shift signals that agent-first labs are re-entering base-model development, betting that small teams with AI-managed experiment pipelines can compete on limited compute. Japanese-language post-training (Namazu) and orchestration (Fugu) offer a template for non-English markets building sovereign AI stacks. Teams weighing whether to build or fine-tune models can benchmark Sakana's three-pillar framing against single-model strategies.

#Sakana AI#Sakana Sawara#Sakana Namazu#Sakana Fugu
官方账号Sakana AI@SakanaAILabsSource