FrogMini-14B-2510: A Specialized Code Debugging Agent
FrogMini-14B-2510, developed by Microsoft Corporation, is a 14 billion parameter language model built upon the Qwen3-14B transformer architecture. Unlike general-purpose LLMs, FrogMini is highly specialized for software engineering tasks, particularly fixing bugs in code within multi-turn debugging workflows. It boasts a substantial 64k token context length, enabling it to handle complex, multi-file debugging scenarios.
Key Capabilities
- Specialized Code Debugging: Fine-tuned on successful debugging trajectories from Claude Sonnet 4 within the BugPilot framework.
- Multi-turn Reasoning: Optimized for iterative debugging processes, learning from realistic debugging patterns across various bug datasets (R2E-Gym, SWE-Smith, FeatAdd).
- Parameter Efficiency: Achieves strong performance on SWE-Bench Verified with a Pass@1 score of 45.3%, demonstrating efficiency for its 14B parameter size compared to larger open-weight models.
- Multi-file Code Repair: Designed to handle complex codebases and resolve feature-level bugs across multiple files.
Good for
- Debugging and repairing code in controlled research environments.
- Automated resolution of software bugs across multi-file repositories.
- Research and development of agentic workflows for software engineering.
FrogMini is restricted to text-based inputs and outputs and is not designed for general-purpose text generation or tasks outside of software engineering. Users should always validate generated code for security and correctness.