Refuel LLM

The world’s first large language model purpose-built for data labeling and enrichment

Refuel LLM is trained on a dataset of 5B+ tokens across a curated set 2500+ data labeling and enrichment tasks from domains such as finance, HR, law and software.

Read our research
“We’re amazed at how well Refuel’s LLMs were able to learn the nuances of our business identity data”
Photo of a man smiling and looking at the camera
Ken Chew
Data Science Lead — Middesk
Photo of a man smiling and looking at the camera

Why choose Refuel LLM

1 |

Superhuman accuracy out of the box

Across a benchmark of 15+ data labeling tasks, Refuel LLM (84.2%) outperforms trained human annotators (80.4%) as well as state of the art LLMs like ChatGPT (81.3%), Gemini (81.5%) and Claude (79.3%).

Never worry about LLM hallucinations

Every single output from Refuel LLM is accompanied by a carefully calibrated confidence score, which estimates the model’s inherent level of confidence in generating the answer. Retry, ensemble or discard outputs based on confidence scores to improve reliability and reduce hallucinations (report).

2 |

3 |

Outperform GPT-4 with 15 min of fine-tuning

While Refuel LLM provides superhuman performance out-of-the-box, it's also highly data-efficient to fine-tune and adapt to new tasks. Outperform GPT-4 with <500 data points and 15 minutes of training. Train and deploy a fine-tuned Refuel LLM for your use case all within Refuel Cloud, without the hassle of infrastructure provisioning and managing auto-scaling.

Try out Refuel LLM yourself

Sign up for Refuel Cloud to get access to Refuel LLM, along with fine-tuning and hands-on support.