agentica-org/DeepCoder-14B-Preview

DeepCoder-14B-Preview is a 14.8 billion parameter code reasoning LLM developed by Agentica, fine-tuned from DeepSeek-R1-Distilled-Qwen-14B. It utilizes distributed reinforcement learning (RLLM) and iterative context lengthening to achieve strong performance on coding tasks, including 60.6% Pass@1 accuracy on LiveCodeBench v5. The model excels at code generation and problem-solving, demonstrating robust long-context reasoning up to 131072 tokens.