org2ai/Wald-26B-A4B
org2ai/Wald-26B-A4B is a 26 billion parameter open decision model, built on Gemma 4 26B-A4B, a mixture-of-experts architecture with approximately 4 billion active parameters per token. This model is uniquely designed to output probabilities for typed options given a state, rather than generating text, making it ideal for automated decision-making systems. It achieves a Decision Index 0.3 score of 62.30 and a JevBench-XL TEST score of 75.3%, demonstrating strong performance in structured decision tasks. Its primary use case is providing fast, probabilistic decision outputs for system integration, with a median processing time of 53 ms in one-pass mode.
Loading preview...
Wald-26B-A4B: An Open Decision Model
Wald-26B-A4B, developed by org2ai, is a 26 billion parameter open decision model based on the Gemma 4 26B-A4B mixture-of-experts architecture. Unlike traditional large language models that generate text, this model is specifically engineered to receive a state and typed options, then return a probability for each option. This design eliminates the need for parsing text outputs, streamlining integration into automated decision systems.
Key Capabilities
- Probabilistic Outputs: Provides direct probabilities for given options, simplifying decision logic.
- High Performance on Decision Benchmarks: Achieves a Decision Index 0.3 score of 62.30 (one-pass, complete public suite) and a JevBench-XL TEST score of 75.3%.
- Fast Inference: Delivers a median processing time of 53 ms in one-pass mode on an RTX PRO 6000 GPU, making it suitable for low-latency applications.
- Optional "Thinking" Mode: An "Auto" mode allows the model to engage Gemma 4's native reasoning when the top probability is below 0.7, enhancing decision accuracy for uncertain cases.
- Open and Deployable: Released under Apache-2.0 license with BF16 weights (52 GB), designed for self-hosting on a single 96 GB GPU.
Good For
- Automated Decision Systems: Ideal for applications requiring structured, probabilistic outputs for choices, classifications, or routing.
- High-Throughput Environments: Its speed and direct probabilistic output are beneficial for systems needing rapid, consistent decisions.
- Jev-Compatible Integrations: Designed to work with Jev-compatible
POST /v1/systemoneAPIs, facilitating seamless integration into existing Jev-based workflows. - Structured Reasoning Tasks: Excels in scenarios where clear, quantifiable decisions are needed rather than open-ended text generation.