startlux-models/StartLux-Decision-v1.1-9B-Instant

Hugging Face
VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 10, 2026License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

StartLux-Decision-v1.1-9B-Instant is a 9 billion parameter decision model developed by StartLux Labs, designed to provide probabilistic answers to typed questions based on a given state, which can include text or up to 8 images. Unlike traditional LLMs, it does not generate text but instead returns probabilities for predefined options (choice, yes/no, score) in a single forward pass. This model is optimized for structured decision-making tasks, offering a unique approach to classification and evaluation by directly outputting confidence scores for specific choices.

Loading preview...

StartLux-Decision-v1.1-9B-Instant Overview

StartLux-Decision-v1.1-9B-Instant is a 9 billion parameter model from StartLux Labs, specifically engineered for structured decision-making. It processes a given state, which can be text or JSON and include up to 8 images, and answers typed questions by providing probabilities for predefined options. This model is distinct from generative LLMs as it directly outputs confidence scores for choices (e.g., choice, yes/no, score) rather than generating free-form text.

Key Capabilities

  • Direct Decision Output: Returns probabilities for options in a single forward pass, eliminating text generation.
  • Multimodal Input: Supports state descriptions with text, JSON, and up to 8 images.
  • Structured Question Types: Handles choice (pick one of several options), noul (yes/no), and score (level on a scale of 2 to 10) questions.
  • Bidirectional Prompt Attention: Utilizes a unique attention mechanism where prompt tokens attend to later prompt tokens, allowing the model to read the state with the question in view.
  • API Compatibility: Request and response formats are compatible with the System One /v1/systemone API.
  • Performance: Achieves a Decision Index 0.3 score of 69.96 and a DecisionPivot pair accuracy of 47.5 for the 9B Instant variant.

Good For

  • Automated Decision Systems: Ideal for applications requiring direct, probabilistic answers to structured questions.
  • Classification and Evaluation: Suitable for tasks where inputs need to be categorized or rated based on specific criteria.
  • Multimodal Analysis: Effective in scenarios where decisions depend on both textual and visual information.
  • High-Throughput Inference: Designed for efficient processing, with low latency for decision read-outs, especially with vLLM.
  • Non-Commercial Research: The weights are available under a CC BY-NC 4.0 license for research and non-commercial use. Commercial use requires a separate license from StartLux Labs.

Finetunes
4 models