Nous Research releases open-source coding model NousCoder-14B, trained in 4 days on 48 B200 GPUs

Nous Research, an open-source AI startup backed by crypto venture firm Paradigm, on Monday released a new competitive coding model that it says matches or outperforms several larger proprietary systems. The model, named NousCoder-14B, was trained in just four days using 48 Nvidia B200 GPUs. It achieves an accuracy of 67.87% on LiveCodeBench v6, a standardized benchmark testing models on competitive programming problems published from August 2024 to May 2025. This score is 7.08 percentage points higher than the base model Qwen3-14B from Alibaba, from which training began. The release comes amid heightened interest in AI coding tools: since the New Year, discussions have been dominated by Anthropic’s agentic tool Claude Code, which developers say can reproduce in hours code that previously took months to create. These concurrent events highlight how rapidly AI-assisted coding is evolving and how fierce the competition among companies for this technology has become.