Hedra Omnia
AI model · Multimodal model · Text → Text by Hedra
A unified base model for post-training and multimodal adaptation.
Hedra Omnia is a unified base model for post-training and multimodal adaptation. It provides a foundation for researchers and enterprises to build specialized visual and physical AI models using Hedra's infrastructure and expertise.
Model specs
The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.
- Launched
- 2026-02-05
Category: Generative AI
- Sizes and variants
- Unified base model
- Access
- Model Factory engagements for custom development
- Released
- 2026-02-05
- Model Type
- Unified base model
- Primary Use Case
- Post-training foundation
- Adaptation Methods
- SFT and RL
- Unified base model for post-training adaptation
- Supports supervised fine-tuning (SFT) and reinforcement learning (RL)
- Foundation for specialized visual and physical AI models
- Released February 5, 2026
Overview
Hedra Omnia is a unified base model designed specifically for post-training research and model adaptation. It serves as a foundation for researchers and enterprises to build specialized multimodal models using supervised fine-tuning and reinforcement learning. The model is part of Hedra's broader research into general visual intelligence and physical AI applications. Hedra provides the necessary research expertise and GPU infrastructure to help organizations adapt the model for their specific needs. Evaluation processes for Omnia-based models include rigorous testing for target behaviors and regressions in video attributes like motion and sharpness. It is typically accessed through Hedra's Model Factory for custom development and deployment.
Model Adaptation and Research
- Unified Base Model: Hedra Omnia serves as a unified starting point for researchers and enterprises to build specialized multimodal models.
- Post-training Foundation: The model is designed specifically to support post-training research and application development.
- Supervised Fine-tuning Support: Hedra Omnia supports adaptation through supervised fine-tuning (SFT) to pair representative inputs with desired outputs.
- Reinforcement Learning Ready: The model architecture allows for reinforcement learning (RL) using rollouts and reward functions to guide updates.
Evaluation and Deployment
- Research Infrastructure: Hedra provides research expertise and GPU infrastructure for organizations to adapt models they own.
- Visual and Physical AI Focus: The model is part of Hedra's research into general visual intelligence and physical AI applications.
- Target Behavior Evaluation: Hedra evaluates checkpoints against the base model to ensure target behaviors are learned and capabilities retained.
- Video Regression Testing: Evaluation processes include checking for regressions in video scene content, motion, and sharpness.
Use cases
- Researchers use Hedra Omnia as a starting point to build specialized multimodal models through supervised fine-tuning.
- Enterprises adapt the base model using reinforcement learning to achieve specific target behaviors for physical AI applications.
- Developers evaluate model checkpoints against the Omnia base to ensure new training hasn't caused regressions in video motion or sharpness.
Specifications
- Sizes and variants
- Unified base model
- Access
- Model Factory engagements for custom development
- Fine-tuning
- Supports supervised fine-tuning (SFT) and reinforcement learning (RL)
- Released
- 2026-02-05
- Model Type
- Unified base model
- Primary Use Case
- Post-training foundation
- Adaptation Methods
- SFT and RL
- Release Date
- 2026-02-05
How it works
- Model Factory Engagement: Enterprises and researchers engage with Hedra's Model Factory to access the Omnia base model for custom development.
- Post-training Adaptation: The model is adapted using supervised fine-tuning on representative input-output pairs or reinforcement learning with reward functions.
- Evaluation and Testing: Checkpoints are compared against the base model to verify target behaviors and check for regressions in video quality or motion.
- Deployment: The specialized model is deployed using options such as sovereign deployment or forward-deployed researcher support.
FAQ
What is Hedra Omnia?
Hedra Omnia is a unified base model designed for post-training and model adaptation, serving as a foundation for specialized multimodal models.
Who is Hedra Omnia for?
It is primarily intended for researchers and enterprises looking to build specialized visual and physical AI models through custom development.
How can Hedra Omnia be customized?
The model supports adaptation through supervised fine-tuning (SFT) and reinforcement learning (RL) to achieve specific target behaviors.
How does Hedra ensure model quality during adaptation?
Hedra evaluates checkpoints for regressions in video content, motion, and sharpness, and performs failure analysis on high-scoring outputs.
Added by AntSpace from public sources (hedra.com).