Skip to content

Steiner-32b-preview

A reasoning model trained on synthetic data

Visit official site

What Steiner-32b-preview is

The model is described as an in-progress reasoning series trained with reinforcement learning on synthetic data, open-sourced to share knowledge and gather feedback from real users.

What you can do with it

  • Run zero-shot reasoning without chain-of-thought prompts
  • Explore several reasoning paths and verify or backtrack
  • Read detailed reasoning steps alongside the result

Application scenarios

  • Check accuracy on GPQA Diamond subfields such as molecular biology
  • Add parameters such as skip_special_tokens false in a request
  • Send a question through a Python client and read the output

What to watch out for

  • The project is stated to be in development
  • Multi-turn dialogue is described as not recommended
  • Full o1-style inference scaling is described as not yet reached

Pros & cons

✓ What we like

  • Detailed reasoning steps and results are provided
  • Existing inference services are supported

! What to watch out for

  • The project is described as still in development
  • Multi-turn dialogue is not recommended for this model

FAQ

Who developed it?

The material names Yichao Peak Ji as the developer of the reasoning series.

Is o1 scaling fully reached?

The material states the reasoning scaling of OpenAI o1 is not yet fully achieved.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review