Nous Research has released NousCoder-14B, a 14B-parameter open-source coding model that the company says reaches 67.87% on LiveCodeBench v6 after a four-day reinforcement learning run on 48 Nvidia B200 GPUs. The launch lands as developer attention is shifting toward more autonomous coding tools, raising questions about whether open models can keep pace with proprietary systems and whether reproducible training stacks will matter as much as benchmark scores.