falloutxvats/Aero-DeepSeek-R1-Distill-Qwen-32B

TEXT GENERATIONPricing:Input $2.72 / Output $4.8Concurrent Unit Cost:2Model Size:32.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 8, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

The falloutxvats/Aero-DeepSeek-R1-Distill-Qwen-32B is a 32.8 billion parameter distilled language model from DeepSeek-AI, based on the Qwen2.5 architecture with a 32768 token context length. It is fine-tuned using reasoning patterns generated by the larger DeepSeek-R1 model, which was developed through large-scale reinforcement learning. This model excels in reasoning, math, and code tasks, demonstrating strong performance on benchmarks like AIME 2024 and MATH-500, and is designed for efficient deployment of advanced reasoning capabilities.

Loading preview...

Overview

This model, falloutxvats/Aero-DeepSeek-R1-Distill-Qwen-32B, is a 32.8 billion parameter language model developed by DeepSeek-AI. It is a distilled version of the larger DeepSeek-R1 model, which was uniquely trained using large-scale reinforcement learning (RL) directly on a base model, without an initial supervised fine-tuning (SFT) step. This RL-first approach allowed DeepSeek-R1 to naturally develop advanced reasoning behaviors like self-verification and generating long chains-of-thought.

Key Capabilities

  • Reasoning Distillation: The model demonstrates that complex reasoning patterns from larger models can be effectively distilled into smaller, more efficient models. It leverages reasoning data generated by DeepSeek-R1 to achieve superior performance compared to models where reasoning patterns are discovered through RL on smaller models directly.
  • Strong Performance: On various benchmarks, this 32.8B parameter model shows competitive results, outperforming models like OpenAI-o1-mini in several categories. For instance, it achieves 72.6% on AIME 2024 pass@1 and 94.3% on MATH-500 pass@1.
  • Optimized for Math and Code: Evaluation results highlight its proficiency in mathematical and coding tasks, with strong scores on AIME, MATH-500, and LiveCodeBench.
  • Qwen2.5 Base: Built upon the Qwen2.5-32B architecture, it benefits from the robust foundation of this widely used model series.

Usage Recommendations

  • Prompting: Avoid system prompts; all instructions should be within the user prompt. For math problems, include a directive like "Please reason step by step, and put your final answer within \boxed{}".
  • Reasoning Enforcement: To ensure thorough reasoning, it's recommended to enforce the model to start its response with "\n".

Finetunes
1 models