unsloth/OpenReasoning-Nemotron-32B

TEXT GENERATIONPricing:Input $2.72 / Output $4.8Concurrent Unit Cost:2Model Size:32.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 21, 2025License:cc-by-4.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The unsloth/OpenReasoning-Nemotron-32B is a 32.8 billion parameter large language model developed by NVIDIA, derived from Qwen2.5-32B-Instruct. It is specifically post-trained for advanced reasoning tasks in mathematics, coding, and science solution generation, supporting a context length of 32768 tokens. This model excels in competitive problem-solving, demonstrating strong performance across challenging reasoning benchmarks and offering enhanced capabilities through a generative solution selection (GenSelect) inference mode.

Loading preview...

OpenReasoning-Nemotron-32B Overview

OpenReasoning-Nemotron-32B is a 32.8 billion parameter large language model developed by NVIDIA, built upon the Qwen2.5-32B-Instruct architecture. It is specifically post-trained to excel in complex reasoning tasks across mathematics, coding, and science. This model is part of a family of OpenReasoning models, available in various sizes (1.5B, 7B, 14B, 32B), all optimized for competitive problem-solving.

Key Capabilities

  • Advanced Reasoning: Demonstrates exceptional performance on challenging reasoning benchmarks, including GPQA, MMLU-PRO, LiveCodeBench, and AIME.
  • Multi-domain Expertise: Specialized in generating solutions for math, code, and science problems.
  • Generative Solution Selection (GenSelect): Can be used in a "heavy" inference mode that combines multiple parallel generations to select the best solution, significantly boosting performance on math and coding benchmarks, even surpassing O3 (High) levels.
  • High Output Token Support: Trained to handle up to 64,000 output tokens, suitable for detailed and extensive problem solutions.

Good For

  • Competitive Programming and Math: Ideal for developers and researchers working on competitive math, coding, and science challenges.
  • Complex Problem Solving: Suited for applications requiring robust reasoning capabilities and the generation of detailed, accurate solutions.
  • Research and Development: Provides a strong foundation for further research in AI reasoning and solution generation, particularly with its GenSelect capabilities.

This model is ready for commercial and non-commercial research use under the Creative Commons Attribution 4.0 International License (CC-BY-4.0).

Finetunes
2 models