Skip to content

MAP-NEO

A fully open language model with data, pipeline and training code

Visit official site
Category AI models

What MAP-NEO is

The model is described as a fully open large language model trained from scratch on 4.5T English and Chinese tokens, released with data, pipeline, scripts and alignment code for full transparency.

What you can do with it

  • Pretraining data and the Matrix data processing pipeline
  • Pretraining scripts and alignment code
  • Intermediate checkpoints, a self-trained tokenizer and the pretraining corpus

Application scenarios

  • Build a chatbot for customer service
  • Integrate into a search engine to improve result relevance
  • Generate abstracts and content for news and articles

Pros & cons

✓ What we like

  • Released under the MIT licence for commercial use
  • Strong results on reasoning, mathematics and coding tasks

! What to watch out for

  • Setup begins with reading the project documentation
  • Input data goes through the Matrix pipeline

FAQ

Who is the release for?

The material names researchers and developers, education institutions and enterprises.

What licence applies?

The material states the MIT licence, which supports commercial use.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review