orca-wm/Orca

Health v2 · maintenance only

Flags: no_releases young no_license

Adoption not part of the score

1038 stars · 58 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Orca is a general world foundation model from BAAI centered on Next-State-Prediction, learning a unified world latent space from visual and language signals. The repository provides inference code and the Orca-4B checkpoint for world latent learning and downstream text/image readouts.

Use cases

• generate predictions of future world states from video and language inputs
• run inference with a pretrained 4B world model checkpoint
• extract world latent representations for downstream tasks
• research unified next-state prediction instead of next-token or next-frame objectives
• train lightweight readout decoders on a frozen world model encoder
• explore scaling behavior of world foundation models

When to choose

• you need a pretrained multimodal world model for research on state prediction
• you want to probe or read out a unified world latent with text and image decoders
• you are studying world-model scaling or unconscious/conscious learning from video

When to avoid

• you need a production-ready application or fine-tuning pipeline (fine-tuning code is not yet released)
• you need a small lightweight model (the 0.8B checkpoint is not yet available)
• you require a permissively licensed project (no license is specified)

Facets

library · maturity experimental

machine-learning deep-learning llm-inference video-processing nlp artificial-intelligence machine-learning deep-learning computer-vision python world-model next-state-prediction foundation-model multimodal world-latent research-code video-understanding natural-language-processing gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="orca-wm/Orca")