microsoft/NextCoder-32B

microsoft/NextCoder-32B is a 32.5 billion parameter causal language model developed by Microsoft, based on the Qwen2.5-Coder Instruct architecture. It is specifically fine-tuned using the novel Selective Knowledge Transfer (SeleKT) methodology for robust code editing. This model demonstrates significant improvements in complex code editing tasks, performing comparably to GPT-4o on benchmarks like Aider-Polyglot, while maintaining generalizability and supporting a long context of up to 32K tokens.