Current location: Home> AI Model> Reinforcement Learning
Step Reasoner mini

Step Reasoner mini

Are you still worried about complex mathematical logic problems and boring text creation? Step R-mini helps you solve it easily!
Author:LoRA
Inclusion Time:16 Jan 2025
Downloads:811
Pricing Model:Free
Version:v1.0
Introduction

Step Reasoner mini (Step R-mini for short) is the first reasoning model launched by Leap Star. It uses a unique "slow thinking" and repeatedly verified logic mechanism to provide accurate and reliable responses, and can effectively solve complex problems such as logical reasoning, coding, mathematics, etc., while taking into account general fields such as literary creation, showing a powerful " Ability to study both arts and sciences.

Core features:

  • Strong reasoning ability: Good at active planning, trial and reflection, and solving complex problems through logical reasoning, including mathematics problems (even Mathematical Olympiad problems), geometry problems (can actively draw sketches), logical reasoning problems and LeetCode "Hard" level programming problems .

  • A combination of arts and sciences: Unlike many inference models that are only good at a single field, Step R-mini has been trained through a large amount of reinforcement learning to perform well in tasks such as literary creation, daily chatting and translation, and can understand user intentions and perform creative tasks. Express.

  • Excellent benchmark performance: In mathematical benchmarks such as AIME and Math, Step R-mini performs better than o1-preview and is comparable to OpenAI's o1-mini; it also performs better than o1-preview in LiveCodeBench programming tasks.

  • Reinforcement learning training: Use the On-Policy reinforcement learning algorithm for training to improve the comprehensive capabilities of the model.

  • Future visual reasoning capabilities: Step Star is developing a visual reasoning model to extend reasoning capabilities to the visual field and achieve "Spatial-Slow-Thinking".

Application scenarios:

  • Math problem solving: Ability to construct chains of reasoning, enumerate solutions, and draw sketches.

  • Logical reasoning: Ability to independently explore problem-solving ideas and self-questioning.

  • Programming: Able to understand user needs and build code logic to solve complex development needs.

  • Content Creation: Able to understand users’ needs and express creatively.

  • Translation: Ability to translate accurately and with rich connotations.

How to experience:

Users can log in to the Yuewen web page https://yuewen.cn , select "Step R-mini" in the upper left corner to experience it.

Preview
Guess you like
  • Step Reasoner mini

    Step Reasoner mini

    Are you still worried about complex mathematical logic problems and boring text creation? Step R-mini helps you solve it easily!
    AI reasoning model a reasoning model with both liberal arts and science capabilities
  • Microsoft Phi-4

    Microsoft Phi-4

    Microsoft Phi-4 is an artificial intelligence (AI) framework developed by Microsoft for automated training and inference of deep learning and reinforcement learning tasks.
    Small language models mathematics
Selected columns
  • ComfyUI

    ComfyUI

    The ComfyUI column provides you with a comprehensive ComfyUI teaching guide, covering detailed tutorials from beginner to advanced, and also collects the latest news ComfyUI , including feature updates, usage skills and community dynamics, to help you quickly master this powerful AI image generation tool!
  • Runway

    Runway

    Explore the infinite possibilities of Runway ai, where we bring together cutting-edge technological insights, practical application cases and in-depth analysis.
  • Cursor

    Cursor

    Cursor uses code generation to debugging skills, and here we provide you with the latest tutorials, practical experience and developer insights to help you with the programming journey.
  • Sora

    Sora

    Get the latest news, creative cases and practical tutorials Sora to help you easily create high-quality video content.
  • Gemini

    Gemini

    From performance analysis to practical cases, we have an in-depth understanding of the technological breakthroughs and application scenarios of Google Gemini AI.