eventvoid/yes-or-no-1.5b

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 18, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The eventvoid/yes-or-no-1.5b is a 1.5 billion parameter language model specifically designed to answer yes/no questions with nuanced confidence levels. It provides exactly one of nine predefined string responses, including 'Maybe' for undetermined or unknowable questions. This model excels at refusing non-questions, ignoring injected instructions, and maintaining an operator policy for caution, making it highly reliable for binary decision-making tasks where clear, constrained outputs are critical.

Loading preview...

Overview

The eventvoid/yes-or-no-1.5b is a 1.5 billion parameter model uniquely engineered to provide precise yes/no answers, along with a 'Maybe' option for ambiguous or unknowable queries. It strictly adheres to a set of nine predefined string responses (e.g., Yes!, No., Maybe?), where punctuation indicates confidence levels (! for high, . for medium, ? for low). This model is designed to be robust against instruction injection and to enforce an operator policy for cautious responses.

Key Capabilities

  • Strict Output Format: Always responds with one of nine specific strings, ensuring predictable and parseable output.
  • Confidence Nuance: Uses punctuation to convey the model's confidence in its answer.
  • 'Maybe' Responses: Explicitly handles undetermined or unknowable questions, preventing forced incorrect answers.
  • Instruction Following: Refuses non-questions and ignores injected instructions, maintaining focus on its core task.
  • High Accuracy on Hazardous Questions: Achieved 96.4% accuracy on dangerous questions, with zero incorrect 'yes' answers.
  • Limited Context: Uses only the system turn and the last user message, with a context window of 2048 tokens.

Limitations

  • Language: Only supports English.
  • Knowledge Cutoff: Not reliable for information after 2023.
  • Specific Tasks: Performs poorly on spelling, arithmetic beyond small numbers, and contested claims.

Good For

  • Applications requiring highly constrained, binary (or ternary with 'Maybe') responses.
  • Use cases where admitting uncertainty is crucial.
  • Filtering or routing based on clear yes/no decisions.
  • Scenarios demanding resistance to prompt injection for simple questions.

Finetunes
1 models