Fine-Tuning a 32B Legal LLM That Outperformed a Frontier Model at 4× Lower Serving Cost

LegML fine-tuned a 32B legal LLM using GRPO and supervised fine-tuning that beat a frontier model on legal benchmarks.