Mapika/decider-2b-vision

VISIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2.3BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 16, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Mapika/decider-2b-vision is a 2.3 billion parameter vision-language model based on Qwen3.5-2B, specifically fine-tuned for making typed decisions from images. It processes an image and a text question with lettered options to yield a calibrated probability over the choices in a single forward pass. This model excels at tasks requiring visual decision-making, such as analyzing game frames or multiple-choice image questions, without generating free-form text.

Loading preview...

decider-2b-vision: Image-Based Decision Making

Mapika/decider-2b-vision is a 2.3 billion parameter vision-language model, built upon the Qwen3.5-2B architecture and incorporating v5 language weights from the decider-2b model. Its core function is to provide calibrated probabilities for lettered options in response to an image and a text question, performing typed decisions in a single forward pass rather than generating free-form text or captions.

Key Capabilities

  • Visual Decision-Making: Processes images (photos, diagrams, game frames) alongside text questions to select from predefined options.
  • Calibrated Probabilities: Outputs a calibrated probability distribution over the given answer choices.
  • Efficient Processing: Designed for rapid decision inference, with a 256x240 game frame costing only 64 visual tokens.
  • Text-Only Support: Retains the text-only decision-making capabilities and abstention handling of the decider-2b v5 model.
  • Performance: Achieves 0.89 accuracy on Visual7W (held out) and 0.96 on Pong/Breakout frames, demonstrating strong performance in visual understanding and game-play scenarios.

Good For

  • Game AI: Analyzing game frames (e.g., Pong, Breakout) to make decisions based on visual input.
  • Visual Question Answering (VQA): Answering multiple-choice questions that require understanding of an image, such as those found in A-OKVQA, AI2D, and ScienceQA datasets.
  • Automated Routing/Classification: Systems requiring rapid, image-informed classification or routing decisions.
  • Structured Output from Visuals: Use cases where a specific, typed decision is needed from visual data, rather than open-ended text generation.

Finetunes
1 models