org2ai/Wald-4B
Wald-4B is a 4.5 billion parameter open decision model developed by org2ai, designed to provide calibrated probabilities for typed options given a state. Built on the Qwen3.5-4B architecture, it features a 32,768 token context length and excels at decision-making tasks, including web actions and zero-shot image and video analysis via Qwen3.5-4B's vision tower. Its primary use case is providing fast, probabilistic decisions for applications requiring structured outputs rather than generative text.
Loading preview...
Wald-4B v2.1: A Specialized Decision Model
Wald-4B v2.1, developed by org2ai, is a 4.5 billion parameter open decision model built upon the Qwen3.5-4B architecture. Unlike traditional LLMs that generate text, Wald-4B is specifically engineered to output calibrated probabilities for a set of typed options based on a given state, making it ideal for structured decision-making tasks. It boasts a 32,768 token context length and is released under an Apache-2.0 license.
Key Capabilities
- Probabilistic Decisions: Provides calibrated probabilities for each option, eliminating the need for text parsing.
- Multimodal Input: Features zero-shot image and video understanding through Qwen3.5-4B's vision tower, enabling decisions based on visual data.
- High Performance: Achieves a Decision Index 0.3 score of 60.05 and a median processing time of 158 ms for serial requests.
- Efficient Decision-Making: "Thinks only when unsure" using an Auto 0.7 mechanism, natively processing about half of requests without additional thought.
- Web Action Proficiency: Demonstrates strong performance in Mind2Web element selection with 92.3% accuracy (6 candidates).
Good For
- Applications requiring structured, probabilistic outputs for decision-making.
- Systems needing fast, calibrated responses to choices or classifications.
- Use cases involving multimodal input, such as analyzing images or video clips to make decisions.
- Integrating into systems where a small, efficient, and open-source decision model is preferred.