nvidia/OpenReasoning-Nemotron-32B

Hugging Face
TEXT GENERATIONConcurrency Cost:2Model Size:32.8BQuant:FP8Ctx Length:32kPublished:Jul 15, 2025License:cc-by-4.0Architecture:Transformer0.1K Open Weights Warm

nvidia/OpenReasoning-Nemotron-32B is a 32.8 billion parameter large language model developed by NVIDIA, derived from Qwen2.5-32B. It is specifically post-trained for advanced reasoning tasks in mathematics, code generation, and science solution generation, supporting up to 64,000 output tokens. This model excels in competitive reasoning benchmarks and can be enhanced with Generative Solution Selection (GenSelect) for improved performance.

Loading preview...

OpenReasoning-Nemotron-32B: Advanced Reasoning Model

OpenReasoning-Nemotron-32B is a 32.8 billion parameter language model developed by NVIDIA, based on the Qwen2.5-32B architecture. It is specifically post-trained to excel in complex reasoning tasks across math, code, and science, supporting an extensive context length for up to 64,000 output tokens.

Key Capabilities

  • Specialized Reasoning: Optimized for generating solutions in competitive math, coding, and scientific problems.
  • High Performance: Demonstrates strong results on challenging reasoning benchmarks such as AIME, LiveCodeBench, GPQA, and MMLU-PRO, often setting new records for its size class.
  • Generative Solution Selection (GenSelect): Incorporates a unique inference mode that combines multiple parallel generations to select the best solution, significantly boosting performance on math and coding benchmarks. This capability generalizes across problem types.
  • Commercial Use: Available for both commercial and non-commercial research under the Creative Commons Attribution 4.0 International License (CC-BY-4.0).

Good For

  • Developers and researchers focused on competitive programming, mathematical problem-solving, and scientific inquiry.
  • Applications requiring robust, step-by-step reasoning and solution generation.
  • Leveraging advanced inference techniques like GenSelect for enhanced accuracy in complex tasks.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p