AI & Robotics
NousCoder-14B Brings Fully Open-Source RL to Competitive Coding
Nous Research releases a 14-billion-parameter coding model trained with reinforcement learning in just four days, publishing the entire training stack.
Key Takeaways
- NousCoder-14B achieves 67.87% on LiveCodeBench v6, a 7-point improvement over its Qwen3-14B base model through reinforcement learning
- The entire training pipeline, RL environment, and benchmark suite are published for full reproducibility by any researcher
- Training took just four days on 48 Nvidia B200 GPUs using Nous Research's Atropos reinforcement learning framework
DE
DT Editorial AI··via venturebeat.com