Nous Research's NousCoder-14B is an open-source coding model landing right in the Claude Code moment
67.87%. That's NousCoder-14B's accuracy on competitive programming—trained in 4 days on 48 GPUs. Open-source just caught up to Claude Code.
Why it matters
Nous Research released a competitive open-source coding model that matches proprietary systems while publishing full training infrastructure, positioning open-source as a viable alternative to closed models in the high-stakes AI coding race dominated by Anthropic's Claude Code.
The key facts
11 to knowNousCoder-14B achieves 67.87% accuracy on LiveCodeBench v6
7.08 percentage point improvement over base model (Qwen3-14B)
Trained in 4 days using 48 NVIDIA B200 GPUs
24,000 competitive programming problems used for training
Model trained from ~1600-1750 Codeforces rating to 2100-2200 equivalent
Context window: 32K initial, 40K training, 80K evaluation
Complete Atropos training stack open-sourced on Hugging Face (Apache 2.0 license)
Nous Research raised $65M total funding (Paradigm-led $50M round in April 2025)
Data scarcity identified: ~24,000 problems represents most high-quality competitive programming data available
Key innovation: DAPO (Dynamic Sampling Policy Optimization) with dynamic sampling and iterative context extension
Released amid viral Claude Code demonstrations (e.g., Google's Jaana Dogan post comparing to 1-year dev effort)
Go to the source
VentureBeat AIventurebeat.com
Publisher excerpt: Nous Research, the open-source artificial intelligence startup backed by crypto venture firm Paradigm, released a new competitive programming model on Monday that it says matches or exceeds several larger proprietary systems — trained in just four days using 48 of Nvidia's latest B200 graphics…