Skip to content

Skywork-o1-Open-PRM-Qwen-2.5-7B

Open PRM models for o1-style slow reasoning

Visit official site

What Skywork-o1-Open-PRM-Qwen-2.5-7B is

The series is described as open reasoning models from the Skywork team that combine o1-style slow thinking with process rewards, showing thinking, planning and reflection in their output.

What you can do with it

  • Produce reasoning steps and rewards
  • Score each step of code
  • Tackle harder reasoning tasks with the 7B model

Application scenarios

  • Generate reasoning steps for math questions
  • Grade code step by step in evaluation
  • Run cross-lingual reasoning on Chinese and English data

Getting started

  • Clone the PRM inference repository
  • Prepare input data and run PRM inference
  • Install the inference packages and plugin
  • Start the inference server for the model

Pros & cons

✓ What we like

  • Multilingual data covers Chinese and English
  • The 7B model extends the 1.5B model to harder tasks

! What to watch out for

  • The described gains are on standard benchmarks
  • Local inference needs a cloned repository and a server

FAQ

Who developed the series?

The material names the Skywork team at Kunlun Tech.

Which datasets are listed?

The material names OlympiadBench, AIME-24 and AMC-23 alongside Chinese and English data.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review