Skip to main content
Hong Kong
AIMenta
P

PhoBERT

by VinAI Research · est. 2020

PhoBERT is VinAI Research's Vietnamese language BERT model, pre-trained on a large Vietnamese corpus. It remains the standard base model for Vietnamese NLP tasks involving classification, named entity recognition, and sentiment analysis — domains where the bidirectional pre-training gives PhoBERT an advantage over general-purpose generative models.

Visit PhoBERT (opens in new tab) Get a recommendation
AIMenta verdict
Niche use
3/5

"The standard Vietnamese BERT model for classification and NER tasks. Open weights, widely used in Vietnamese NLP research. Best for short-text Vietnamese classification where PhoBERT's pretraining gives it an edge over general multilingual models."

Features
5
Use cases
4
Watch outs
4
What it does

Key features

  • Vietnamese-specific BERT pretraining (large Vietnamese corpus)
  • Two sizes: base and large
  • Apache 2.0 licence (full commercial use)
  • Strong Vietnamese NER and classification performance
  • PyTorch and HuggingFace compatible
When to reach for it

Best for

  • Vietnamese named entity recognition (person, organisation, location extraction)
  • Vietnamese sentiment analysis
  • Short-text Vietnamese classification tasks
  • Academic and research Vietnamese NLP baselines
Don't get burned

Limitations to know

  • ! Encoder-only (BERT-style) — cannot generate text, only encode/classify
  • ! Limited to Vietnamese — not multilingual
  • ! Does not handle long documents well (512 token limit)
  • ! Less actively maintained than SEA-LION or BGE-M3
Context

About PhoBERT

PhoBERT is a AI productivity tool from VinAI Research, launched in 2020. PhoBERT is VinAI Research's Vietnamese language BERT model, pre-trained on a large Vietnamese corpus. It remains the standard base model for Vietnamese NLP tasks involving classification, named entity recognition, and sentiment analysis — domains where the bidirectional pre-training gives PhoBERT an advantage over general-purpose generative models.

Notable capabilities include Vietnamese-specific BERT pretraining (large Vietnamese corpus), Two sizes: base and large, and Apache 2.0 licence (full commercial use). Teams typically deploy PhoBERT for vietnamese named entity recognition (person, organisation, location extraction) and vietnamese sentiment analysis.

Common trade-offs to weigh: encoder-only (BERT-style) — cannot generate text, only encode/classify and limited to Vietnamese — not multilingual. AIMenta editorial take for APAC mid-market: The standard Vietnamese BERT model for classification and NER tasks. Open weights, widely used in Vietnamese NLP research. Best for short-text Vietnamese classification where PhoBERT's pretraining gives it an edge over general multilingual models.

Beyond this tool

Where this category meets practice depth.

A tool only matters in context. Browse the service pillars that operationalise it, the industries where it ships, and the Asian markets where AIMenta runs adoption programs.

Or browse All tools · Encyclopedia · Case studies · Rankings