Skip to content

Starling-7B

An open LLM trained with RLAIF

Visit official site

What Starling-7B is

The model is described as an open LLM trained with reinforcement learning from AI feedback and the Nectar ranking dataset.

What you can do with it

  • Use the released ranking dataset
  • Load the released reward and language models
  • Try the online demo in LMSYS Chatbot Arena

Model notes

  • MT Bench is listed at 8.09 with GPT-4 as judge
  • Assets are published on HuggingFace
  • The stated aim improves LLM usability and safety

Pros & cons

✓ What we like

  • Starling-LM-7B-alpha is published as a language model
  • The stated goal covers LLM usability and safety

! What to watch out for

  • Training relies on AI feedback signals
  • The ranking labels come from GPT-4

FAQ

What score is listed?

The card records 8.09 on MT Bench with GPT-4 as judge.

What was released?

Nectar, Starling-RM-7B-alpha and Starling-LM-7B-alpha are named.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review