startlux-models/StartLux-Decision-4B

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 29, 2026License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

StartLux-Decision-4B by startlux-models is a 4.5 billion parameter multimodal decision model designed to answer typed questions about a given state, providing probabilities for each option. It supports text, JSON, or image inputs up to 262,144 tokens and is optimized for fast inference, achieving 26.0 ms latency for three questions on an H200 GPU. This model excels at structured decision-making tasks like classification, yes/no questions, and rating on a scale, using the TypeSafe /v1/systemone format.

Loading preview...

StartLux-Decision-4B Overview

StartLux-Decision-4B is a 4.5 billion parameter multimodal model developed by startlux-models, specifically engineered for structured decision-making. It processes diverse inputs including text, JSON, and images, with a substantial context length of up to 262,144 tokens. The model provides probabilistic answers for various question types: selecting from options, binary yes/no, or rating on a scale. It utilizes the TypeSafe /v1/systemone format for requests and responses, ensuring compatibility with existing clients.

Key Capabilities

  • Multimodal Input Processing: Accepts text, JSON, and images as input, allowing for rich contextual understanding.
  • Probabilistic Decision Outputs: For every question, the model returns a probability for each possible option, enabling nuanced decision analysis.
  • High Throughput & Low Latency: Optimized for fast inference, achieving 26.0 ms for three questions and 14.7 ms for a single yes/no question on an H200 GPU, leveraging custom fast kernels and CUDA graph replays.
  • Extended Context Window: Supports long inputs up to 262,144 tokens, beneficial for complex scenarios requiring extensive context.
  • Batch Inference: Includes decide_batch for efficient processing of multiple requests simultaneously.

Good For

  • Automated Triage and Classification: Ideal for categorizing customer support tickets, documents, or other data based on defined criteria.
  • Structured Question Answering: Effectively answers questions requiring a choice, a binary decision, or a rating from a predefined scale.
  • Multimodal Analysis: Suitable for applications where decisions need to be made based on a combination of textual and visual evidence.
  • High-Performance Inference: Applications requiring rapid, low-latency decision outputs in production environments.

Finetunes
4 models