Skip to content

XVERSE-MoE-A36B

A multilingual MoE model with 36B active parameters

Visit official site
Category AI models

What XVERSE-MoE-A36B is

XVERSE-MoE-A36B is described as a multilingual large language model built on a mixture-of-experts architecture by XVERSE.

What you can do with it

  • Generate text across more than 40 languages
  • Load the model and tokenizer with Transformers
  • Use fine-tuned data sampling for training

Pros & cons

✓ What we like

  • Fine-grained data sampling and dynamic data switching
  • Custom MoE training framework for higher throughput

! What to watch out for

  • The named training length is 8K samples
  • Loading relies on the Transformers library

FAQ

How many languages are supported?

The material states over 40 languages are supported.

What is the parameter scale?

The material lists 255.4 billion total and 36 billion active parameters.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review