NVIDIA Open Source Code Inference Model (32B, 14B, 7B)

NVIDIA continues to use open source Open Code Reasoning (OCR) Model Kit – Three high-performance large language models for code reasoning and problem solving. 32B, 14B and 7B variants, all released Apache 2.0 License.

Standard to beat the best

this Open code reasoning (OCR) model is included Famous benchmark achievementpoor performance O3-Mini and O1 (low) of Openai Model livecodebench Benchmark. LiveCodeBench is a comprehensive evaluation suite for code inference tasks such as debugging, code generation, and logical completion in real-world developer environments. In direct comparison, NVIDIA’s 32B OCR model ranks first in the reasoning capabilities of open models.

This leap in performance is attributed not only to the model architecture but also to Nvidia Customize “OCR Dataset” – A high-quality code-centric training corpus designed to emphasize tracking, reasoning and multi-step code problem solutions. According to Nvidia, this leads to Token efficiency is 30% higherallowing the model to produce accurate code and logical output with fewer tokens.

Model lineup for each use case

Open code reasoning suite has Three parameter scale:

• OpenCodereasoning-Nemotron-32B
• OpenCodereasoning-Nemotron-14b
• OpenCodereasoning-Nemotron-7b

Each model balances scale and performance. The 32B variant provides the latest results for high-performance inference and research; the 14b model provides low computational requirements and provides powerful inference capabilities, while the 7B variant is ideal for resource-constrained environments while retaining competitive performance on the benchmark.

All models are used Nemotron architecture,NVIDIA’s transformer-based main chain optimizes multi-task learning for multilingual. The weight and configuration of the model are available on the hug surface:

Compatible with open reasoning ecosystems

The key features of these models are Out-of-the-box compatibility With popular reasoning frameworks:

• llama.cpp For lightweight CPU/GPU reasoning
• vLLM GPU services and speculative decoding for optimization
• Transformers By embracing the face of training and evaluating pipeline
• TGI (Text Generation Inference) Extensible API Deployment

This flexibility allows developers, researchers, and businesses to insert these models into existing code AI infrastructure with minimal overhead.

A step to open up code intelligence

With this release, NVIDIA has made a significant contribution to the growing ecosystem of open code models. By positioning Code reasoning – A field that historically dominated by proprietary models – and released under a fully open and relaxed license, NVIDIA empowers the broader AI and developer community to build, fine-tune and deploy advanced inference models in production.

The Open Code Inference Suite adds NVIDIA’s growing portfolio of Open LLMS and strengthens its stance on accessible, transparent AI development. Whether you are a construction developer co-pilot, an automatic code audit agent, or a code generation service, these models offer high-performance, cost-effective, and community-friendly alternatives to closed solutions.

View variants of 32B model, 14B model, 7B model and 32B instruction adjustments. Also, don’t forget to follow us twitter.

Here is a brief overview of what we built in Marktechpost: