launch/MET-D-Qwen3-8B

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 8, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

MET-D-Qwen3-8B is an 8 billion parameter multilingual moral reasoning model fine-tuned from Qwen3-8B. It is designed to judge the acceptability of actions from a specific character's perspective within moral dilemmas, providing a chain-of-thought explanation. The model supports six languages (English, Spanish, Hindi, Korean, Malay, Chinese) and generates both reasoning traces and final answers in the prompt's language. Its primary use is for nuanced, character-perspective moral reasoning across multiple languages.

Loading preview...

Model Overview

MET-D-Qwen3-8B is a specialized 8 billion parameter multilingual moral reasoning model, fine-tuned from the Qwen3-8B base model. Its core function is to evaluate a candidate action within a moral dilemma from a specified character's perspective. The model provides a detailed chain-of-thought explanation before delivering its judgment.

Key Capabilities

  • Moral Reasoning: Judges actions based on a given situation and character description.
  • Perspective-Taking: Answers two specific questions from the character's viewpoint:
    • Is the action acceptable? (Yes / No / Ambiguous)
    • Would doing/not doing the action cause emotional/mental discomfort? (Yes / No)
  • Multilingual Support: Trained on and capable of generating reasoning and answers in six languages: English, Spanish, Hindi, Korean, Malay, and Chinese. This specific checkpoint combines all six languages.
  • Reasoning Trace Generation: Explains its judgment with an explicit chain-of-thought, which is rejection-sampled against ground truth per character perspective.

Training and Uniqueness

The model was trained using self-generated reasoning traces, rejection-sampled to align with ground-truth answers for specific character perspectives and theoretical grounds. This approach addresses the complexity of verifying moral reasoning by conditioning on a per-language, per-situation selection of theoretical grounds. This allows for legible reasoning in the user's native language, making it distinct from models that might only provide English-centric reasoning.

Use Cases

This model is particularly well-suited for applications requiring nuanced moral judgment, ethical AI development, and cross-cultural studies of moral reasoning, especially where character-specific perspectives and detailed explanations in multiple languages are crucial.