KyrlG/NousCoder-14B
TEXT GENERATIONPricing:Input $0.48 / Output $0.96Concurrent Unit Cost:1Model Size:14BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 2, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold
NousCoder-14B is a 14 billion parameter competitive programming model developed by KyrlG and Nous Research. Post-trained on Qwen3-14B using reinforcement learning, it achieves a Pass@1 accuracy of 67.87% on LiveCodeBench v6. This model is specifically optimized for solving complex coding problems and excels in competitive programming scenarios.
Loading preview...
NousCoder-14B: A Competitive Programming Model
NousCoder-14B is a 14 billion parameter model developed by KyrlG and Nous Research, specifically engineered for competitive programming tasks. It is post-trained on the Qwen3-14B architecture using reinforcement learning techniques.
Key Capabilities
- Enhanced Coding Performance: Achieves a Pass@1 accuracy of 67.87% on LiveCodeBench v6, representing a 7.08% improvement over the Qwen3-14B baseline.
- Reinforcement Learning Optimization: Benefits from training on 24,000 verifiable coding problems, utilizing 48 B200 GPUs over four days.
- Large Context Window: Supports a context length of 32768 tokens, allowing for processing of extensive codebases and problem descriptions.
Good For
- Competitive Programming: Excels at solving complex algorithmic challenges and coding puzzles typically found in competitive programming contests.
- Code Generation: Generates high-quality, verifiable code solutions.
- Developer Tools: Can be integrated into tools requiring advanced code understanding and generation capabilities.