KUALA LUMPUR: Nous Research has launched NousCoder-14B, a new competitive AI programming model, aiming to rival larger proprietary systems. This model, trained in just four days using Nvidia's B200 graphics processors, marks a significant step in the rapidly evolving field of AI-assisted software development. It comes at a time when Claude Code, a tool from rival Anthropic, is gaining significant attention for its capabilities. NousCoder-14B is an open-source model that achieves a 67.87 percent accuracy rate on LiveCodeBench v6.
This represents a 7.08 percentage point improvement compared to its predecessor, Qwen3-14B from Alibaba. The release highlights the competitive nature of AI coding tools and the importance of transparency, with Nous Research providing the complete reinforcement learning environment for others to replicate or build upon. "I gave Claude Code a description of the problem, it generated what we built last year in an hour," Jaana Dogan wrote, highlighting the impressive capabilities of the Claude Code tool. Dogan's experience underscores the growing impact AI coding tools have on software development. The release of NousCoder-14B signifies a potential shift towards open-source solutions in AI development, providing infrastructure for reproducible research.
As AI coding models continue to advance, the industry may see increased collaboration and transparency, potentially altering how software is developed across the globe.
Image credit: VENTURE BEAT AI
