Skip to content

Reflection Llama-3.1 70B

An open model that reflects on its own reasoning

Visit official site
Category AI models

What Reflection Llama-3.1 70B is

Reflection Llama-3.1 70B is described as an open LLM trained with Reflection-Tuning on synthetic data from Glaive to catch and fix reasoning errors.

What you can do with it

  • Show inner reasoning with thinking tags
  • Trigger self-correction with reflection tags
  • Sample with standard Llama 3.1 code

Getting started

  • Find the model on Hugging Face and read its docs
  • Set the recommended temperature and top_p
  • Add a Think carefully instruction to the input

Pros & cons

✓ What we like

  • Reflection tags allow self-correction
  • Standard Llama 3.1 sampling code can be reused

! What to watch out for

  • Recommended sampling settings are specified
  • The chat template includes added special tokens

FAQ

What technique is used?

The material names Reflection-Tuning, which lets the model detect and fix reasoning errors.

Which tags appear?

The material lists thinking, closing thinking and reflection tags.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review