contextboxai/Can-1.0-31B

VISIONPricing:Input $0.48 / Cached $0.1 / Output $1.44Concurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 8, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Cân 1.0 (31B) by ContextBox AI is a 31 billion parameter decision model based on Google's Gemma-4-31B-it, designed to provide calibrated probability distributions over supplied options in a single forward pass. It excels at structured decision-making tasks, including choice, scoring, and true/false evaluations, by processing text, JSON, or conversational states. This model is optimized for applications requiring precise, probabilistic decision outputs rather than generative text, making it suitable for rule-based systems, legal analysis, and code evaluation.

Loading preview...

Cân 1.0 (31B): A Specialized Decision Model

Cân 1.0 (31B) is a 31 billion parameter model developed by ContextBox AI, specifically engineered as a decision model. Unlike traditional generative LLMs, Cân processes a given state (text, JSON, or conversation) and a typed question to return a calibrated probability distribution over predefined options in a single forward pass, without generating new text. This makes it highly efficient and precise for structured decision-making.

Key Capabilities

  • Probabilistic Decision Outputs: Provides calibrated probabilities for choice, score (ordered scale), and noul (true/false) question types.
  • Single Forward Pass: Delivers decisions efficiently without text generation, ideal for low-latency applications.
  • Robust Training: Fine-tuned on a diverse dataset including human-labelled decisions, procedural rule-application tasks, synthetic workflows, and programmatically generated math/code items.
  • Teacher Signal Integration: Utilizes soft labels from perplexity-ai/pplx-decider-v1.1-27b mixed with gold labels for enhanced accuracy.
  • Context Handling: Supports states up to 10,000 tokens read in full on 1x H100, with longer states truncated intelligently.

Good For

  • Rule-Based Systems: Automating decisions based on complex rules and criteria.
  • Legal and Compliance Analysis: Evaluating legal documents or regulations to determine outcomes.
  • Code Evaluation: Assessing code correctness or adherence to standards with execution-verified labels.
  • Structured Data Interpretation: Extracting and acting upon decisions from structured or semi-structured data.
  • Applications Requiring Calibrated Probabilities: Where confidence scores for decisions are critical, such as risk assessment or diagnostic tools.

Finetunes
1 models