resect-ai/veritas-8B-fact-checker-non-thinking-1.0
Veritas-8B-Fact-Checker-Non-Thinking-1.0 by Resect Research Labs is an 8 billion parameter model built on the Qwen3 architecture, specifically fine-tuned for fact-checking and factual consistency verification. With a context length of 32768 tokens, it achieves an average balanced accuracy of 75.47% on the LLM-AggreFact benchmark, demonstrating a 2.3% improvement over its base Qwen3-8B model in non-thinking mode. This model is optimized for binary classification tasks to verify claims against provided documents.
Loading preview...
Veritas-8B-Fact-Checker-Non-Thinking-1.0 Overview
Developed by Resect Research Labs, Veritas-8B-Fact-Checker-Non-Thinking-1.0 is an 8 billion parameter model based on the Qwen3 architecture (specifically, Qwen/Qwen3-8B). It has been specialized and fine-tuned using proprietary reinforcement learning and novel techniques to enhance factual grounding and reduce hallucinations in AI models.
Key Capabilities
- Fact-Checking and Factual Consistency Verification: The model excels at determining the factual consistency of claims against provided documents.
- Binary Classification: Designed for a prescribed scoring mode that generates a binary classification (true/false) based on a specified template.
- Improved Performance: Achieves an average balanced accuracy of 75.47% on the LLM-AggreFact benchmark, which aggregates 11 human-annotated datasets. This represents a 2.3% improvement over the base Qwen3-8B model in its non-thinking mode.
Good For
- Automated Fact Verification: Ideal for applications requiring automated assessment of factual claims.
- Enhancing AI Grounding: Useful for integrating into systems that need to detect and mitigate hallucinations by verifying information.
- Integration with MiniCheck: Designed to work seamlessly with the MiniCheck library, providing a straightforward method for scoring documents and claims.