startlux-models/StartLux-Decision-v1.1-27B-Instant

Hugging Face
VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 11, 2026License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

StartLux-Decision-v1.1-27B-Instant is a 27 billion parameter decision model developed by StartLux Labs, designed to provide probabilistic answers to typed questions (choice, yes/no, score) based on a given state, which can include text or up to 8 images. Unlike traditional LLMs, it returns probabilities for all options in a single forward pass without generating text, making it highly efficient for structured decision-making tasks. It supports a 32768 token context length and is optimized for rapid, non-generative decision inference, achieving strong performance on the Decision Index 0.3 benchmark.

Loading preview...

StartLux-Decision-v1.1-27B-Instant Overview

StartLux-Decision-v1.1-27B-Instant is a 27 billion parameter model from StartLux Labs, specifically engineered for structured decision-making rather than text generation. It processes a given state (text or up to 8 images) and a set of typed questions (choice, yes/no, score), returning probabilities for all possible options in a single forward pass. This approach allows for efficient and direct probabilistic outputs without generating conversational text.

Key Capabilities

  • Direct Probabilistic Decisions: Provides probabilities for all options of a given question in one pass, ideal for classification and scoring tasks.
  • Multimodal Input: Accepts both text and up to 8 images as part of the input state for comprehensive context.
  • Efficient Inference: Designed for speed, with median latencies as low as 52ms for 256 tokens on vLLM BF16, and compatible with the System One /v1/systemone API.
  • Specialized Question Types: Handles choice, noul (yes/no), and score (2-10 scale) question formats.
  • 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.

Good For

  • Automated Triage and Routing: Categorizing customer support tickets, assigning severity, or directing inquiries.
  • Structured Data Analysis: Extracting specific decisions or scores from complex textual or visual information.
  • Applications Requiring Fast, Non-Generative Decisions: Use cases where a direct probabilistic answer is needed without the overhead of text generation.
  • Integration with Existing Systems: Its API compatibility with System One facilitates straightforward integration into decision-making workflows.

Finetunes
4 models