Nous Research's NousCoder-14B is an open-source coding model landing right in the Claude Code moment

What we drew from this source

The claims Via News extracted from this document. We point to the source; we don't replace it.

• NousCoder-14B achieves 67.87% accuracy on LiveCodeBench v6, a 7.08 percentage point improvement over base model Qwen3-14B

  80% confidence

• Open-sourcing the Atropos stack provides necessary infrastructure for reproducible olympiad-level reasoning research

  80% confidence

• Claude Code replicated a distributed agent orchestration system her team spent a year developing from a three-paragraph prompt in approximately one hour

  80% confidence

• NousCoder-14B matches or exceeds several larger proprietary systems and was trained in just four days using 48 Nvidia B200 GPUs

  80% confidence

• Within the competitive programming domain, training data limits have been approached

  80% confidence

• Training data is increasingly finite and key future research areas are synthetic data generation and data-efficient algorithms

  80% confidence

• There exists a significant gap in LLM capabilities in creative problem generation compared to humans

  80% confidence

• Humans remain dramatically more sample-efficient learners: Li solved ~1,000 problems over two years while the model required 24,000

  80% confidence

• Once synthetic problem generation is solved, self-play becomes a very interesting direction for AI coding model development

  80% confidence

• NousCoder-14B's improvement trajectory mirrored Joe Li's own two-year journey on Codeforces, accomplished in four days

  80% confidence

Cited in these Via News reports

• Open-Source AI and Developer Tools Are Rewriting Software Engineering Globally →