Step R-mini - The first reasoning model in the Step series launched by Step Star.

What is Step R-mini?

Step R-mini (full name Step Reasoner mini) is a reasoning model launched by Step Star, and the first reasoning model in the Step series. It excels at proactive planning, trial and error, and reflection, providing accurate and reliable responses based on a slow-thinking and iterative verification logic mechanism. The model is adept at solving complex problems such as logical reasoning, coding, and mathematics, and can also handle general domains such as literary creation. Step R-mini performs exceptionally well on mathematical benchmarks and coding tasks, achieving a balance between science and humanities. Step R-mini adheres to the Scaling Law principle, including reinforcement learning, data quality, test-time computation, and model scaling.

Main functions of Step R-mini

• Mathematical problemsConstruct a logical reasoning chain to plan and solve complex mathematical problems step by step. When solving challenging math olympiad problems, enumerate different solution schemes for cross-validation. When dealing with geometry problems, proactively use sketching to build a medium for deep thinking, comprehensively and rigorously analyze the problem requirements, select the best solution formula, and determine whether there are any overlooked factors based on repeated self-questioning.
• Logical reasoningTry different problem-solving approaches independently. After obtaining a preliminary answer, ask yourself if there are other possibilities. Make sure to enumerate all effective solutions. Before submitting the paper, check for any omissions and provide comprehensive and accurate reasoning results.
• Code SolutionIt can correctly solve challenging algorithm problems based on long inference chains, such as problems rated "Hard" on the LeetCode platform. It can also handle complex development requirements, progressively analyzing user needs and intentions, constructing code logic, and interspersing analysis and verification of the current code snippet during code writing, ultimately providing executable code.
• Literary creationTo deeply understand users' expressive needs, analyze creative themes and literary requirements, consider creative perspectives, depicted scenery, rhetorical devices, content structure, etc., endow things with symbolic meaning on the level of human emotions, and add personalized and innovative expressive styles, like a creator who "pursues perfection".

Step R-mini's technological advantages

• Adhere to the Scaling Law principle:
  • Scaling Reinforcement LearningFrom imitation learning to reinforcement learning, from human preferences to environmental feedback, reinforcement learning is used as the core training stage for model iteration.
  • Scaling Data QualityWhile ensuring data quality, we will continue to expand the distribution and scale of data to provide support for enhanced learning and training.
  • Scaling Test-Time ComputeWhile taking into account computational expansion during the testing phase, the System 2 paradigm enables Step-Reasoner mini to perform deep thinking with 50,000 tokens in extremely complex task inference.
  • Scaling Model SizeSystem-2 maintains that scaling up the model is at its core and is developing a more intelligent, general, and comprehensive Step Reasoner reasoning model.
• A balanced education in both arts and sciencesOn math benchmarks such as AIME and Math, it outperforms o1-preview and rivals OpenAI o1-mini. On LiveCodeBench coding tasks, it outperforms o1-preview. Most inference models struggle to balance abilities in both the humanities and sciences; Step R-mini, based on large-scale reinforcement learning training and using an On-Policy reinforcement learning algorithm, achieves a balance between the two.

Step R-mini's project address

• Project official websiteStep R-mini

Step R-mini example demonstration

• Logical reasoning:When tackling logical reasoning tasks, Step R-mini autonomously tries various problem-solving approaches. After obtaining a preliminary answer, it asks itself if there are other possibilities, ensuring that all effective solutions are enumerated, and checks for omissions before submitting the paper.

Application scenarios of Step R-mini

• Educational guidanceIt helps students solve difficult math problems and programming difficulties by providing problem-solving ideas and code examples to help them improve their learning.
• Scientific research supportIt helps researchers perform logical reasoning and data analysis, integrate interdisciplinary knowledge, and promote the progress of research projects.
• Corporate OfficeIt assists programmers in developing code efficiently, provides managers with logical analysis and suggestions for business decisions, and optimizes office processes.
• Literary creationTo inspire cultural and creative workers, provide personalized and innovative literary creation solutions, and enrich the content of works.
• Translation servicesTo meet the demand for high-quality translation, accurately convert languages, and promote cultural exchange and dissemination.