CAST ObjectGen

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit

Not collected

Maximum input tokens

Not collected

Maximum output

Not collected

Reasoning

Not collected

Tool calling

Not collected

Structured output

Not collected

Streaming

Not collected

Batch processing

Not collected

Fine-tuning

Not collected

Knowledge cutoff

Not collected

Release date

Not collected

Publication or observed application version does not establish an original weight release date.

Parameters

1.5 billion total for the VAE and LDM modules, as stated in Section 6.1.Source for Parameters (opens in a new tab)

Do not interpret this as a separately reported DiT-only count.

Architecture

24-layer transformer VAE/LDM geometry system with DINOv2 image features and cross-attention partial-point-cloud conditioning.Source for Architecture (opens in a new tab)

Languages

Not collected

Context caching

Not collected

Usage metering

Not collected

No price or currently callable endpoint for this exact research edition verified. Current Rodin API rates are not inherited.

Data retention

Not collected

No hosting, retention or enterprise contract verified for this research edition.

Input modalities

Masked image, Canonical partial point cloudSource for Input modalities (opens in a new tab)

Output modalities

Object geometry latent, 3D object mesh through VAE decodingSource for Output modalities (opens in a new tab)

Training sequence length

2048 partial-cloud points sampled with farthest-point sampling for conditioning.Source for Training sequence length (opens in a new tab)

Geometric points, not text tokens.

Published training data

Approximately 500,000 filtered Objaverse assets for pretraining; 200,000 curated assets for conditioning, with 32 rendered views per object.Source for Published training data (opens in a new tab)

Training hardware

Conditioning module: 3000 epochs on 64 NVIDIA A800 GPUs, approximately one week; AdamW learning rate 1e-5.Source for Training hardware (opens in a new tab)

Generation modes

Alternates ObjectGen and AlignGen so image-only initial geometry gains progressively stronger canonical point-cloud guidance.Source for Generation modes (opens in a new tab)

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

Why it is included

Qualified unreleased-weight classification as reviewed: the official project page does not provide a released weight download, and reviewed provider repository/model listings do not establish one. The paper documents trained models. This does not assert a permanent proprietary license or rule out future releases; public research publication and integration code are not weight releases.

Lifecycle dates

Sources reviewed