Show HN: Openfusion - enhanced results from a panel of models (github.com)

Openfusion has been launched as an open-source compound-model proxy that enables enhanced responses from a panel of language models (LLMs). Users can direct any OpenAI-compatible tool to Openfusion by specifying "openfusion" as the model. The system takes user prompts, which it distributes across multiple LLMs simultaneously. A designated "judge" model then analyzes the responses, identifying consensus, contradictions, and unique insights to produce a single, coherent answer. This approach represents a transparent alternative to proprietary solutions like OpenRouter's Fusion, providing customization and control over the process without the typical black-box limitations. The significance of Openfusion lies in its ability to improve the quality of AI-generated responses by leveraging multiple models concurrently, which can lead to more accurate and nuanced answers, especially for complex queries. Key technical features include a user-friendly interface for interaction, a customizable YAML configuration for various settings (like panel composition and response aggregation), and built-in tools for benchmarking performance. While initial evaluations indicate that fusion may not outperform single models on straightforward tasks, it shows promise for more challenging prompts where drawing from diverse model perspectives could yield superior responses. This development may encourage deeper exploration into the efficacy of mixture-of-agents methodologies within the AI/ML community.