Gen-Verse/ReasonFlux-F1-7B

ReasonFlux-F1-7B: Advanced Reasoning LLM

ReasonFlux-F1-7B is a 7.6 billion parameter language model developed by Gen-Verse, specifically fine-tuned for state-of-the-art reasoning capabilities. It leverages a novel template-augmented reasoning paradigm, building upon the methodologies introduced in the ReasonFlux-Zero project.

Key Capabilities & Performance

This model is designed to excel in complex reasoning tasks, particularly in mathematics and general problem-solving. While ReasonFlux-F1-7B is the 7B variant, its 32B counterpart, ReasonFlux-F1-32B, demonstrates superior performance against other models in its class on challenging benchmarks:

• MATH500: Achieves 96.0% (ReasonFlux-F1-32B)
• AIME 2024: Achieves 76.7% (ReasonFlux-F1-32B)
• AIME 2025: Achieves 53.3% (ReasonFlux-F1-32B)
• GPQA-Diamond: Achieves 67.2% (ReasonFlux-F1-32B)

These results highlight the effectiveness of the template-augmented reasoning approach in enhancing logical inference and problem-solving accuracy.

Training and Resources

ReasonFlux-F1-7B was fine-tuned using template-augmented reasoning trajectories. The associated research paper, "ReasonFlux: Hierarchical LLM Reasoning via Scaling Thought Templates", provides further details on the methodology. The model's training dataset, Gen-Verse/ReasonFlux-F1-SFT, is publicly available.

Use Cases

ReasonFlux-F1-7B is particularly well-suited for applications requiring strong logical reasoning, mathematical problem-solving, and complex question answering. Its focus on reasoning makes it a valuable tool for tasks where accurate and structured thought processes are critical.