PhoGPT

Open-source Vietnamese large language model from VinAI Research, pre-trained from scratch on Vietnamese text.

LLMs & Chat Open Source Open Source
Researched · Published · Reviewed
RECATOOLS Score
6.5 / 10
HQ
Hanoi, Vietnam
Launched
2023
Developer
VinAI Research (Vingroup)

Overview

PhoGPT is an open-source Vietnamese generative LLM series from VinAI Research, the AI lab of Vietnam's Vingroup. The 7.5B-parameter PhoGPT-7B5 is a Transformer decoder pre-trained from scratch on a large Vietnamese corpus, released alongside an instruction-tuned PhoGPT-7B5-Instruct variant fine-tuned on 150K Vietnamese prompt-response pairs. VinAI has since released lighter PhoGPT-4B models. All variants are published on Hugging Face for open use.

Advertisement

Pricing

Pricing shown for reference only. These figures reflect RECATOOLS research as of 16 Sep 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.

Free
Free
Open model weights free to download from Hugging Face.

What you can produce with PhoGPT

  • Run PhoGPT-7B5, a 7.5B Vietnamese decoder pre-trained from scratch on Vietnamese text
  • Work in Vietnamese without the tokenisation penalty a mostly-English model imposes
  • Self-host rather than routing Vietnamese text through a foreign API
  • Fine-tune it on domain data in Vietnamese
  • Build on VinAI Research, the AI lab of Vingroup
  • Read the practitioner signal on the VinAIResearch/PhoGPT repo — there is no commercial review footprint
Advertisement

ASEAN Perspective

PhoGPT in Southeast Asia

PhoGPT is the Vietnamese model with the clearest practitioner footprint — VinAI Research, Vingroup's lab, pre-trained the 7.5B PhoGPT-7B5 from scratch on Vietnamese rather than adapting an English model, and published it openly with an instruction-tuned variant and lighter 4B versions. It serves as the reference baseline for Vietnamese NLP work. For the local ecosystem, an open baseline to measure against is more useful than a closed, market-leading model. Note the corporate context when planning around it: VinAI's ownership has changed hands, so the lab behind the weights is not the independent research group it was at release.

RECATOOLS Verdict

What this is for: A monolingual Vietnamese LLM pre-trained from scratch for Vietnamese text generation and instruction following.

Who this is for: Vietnamese developers and researchers building local-language chat and NLP applications on open weights.

Availability: Free and open-source on Hugging Face; self-hosted, with no official managed consumer product.

Independent AI-assisted assessment by RECATOOLS.

What people say

PhoGPT has no commercial review footprint, but as an open Vietnamese model it has the clearest practitioner signal of this group. The VinAIResearch/PhoGPT repo is public, the accompanying paper is on arXiv, and the model is widely referenced as a baseline in Vietnamese NLP resource lists — it's routinely treated as the reference open Vietnamese LLM for low-resource research.

On evaluation, VinAI reported human preference tests using a Vietnamese-translated Vicuna benchmark (80 questions across 8 categories), with PhoGPT-7B5-Instruct compared against ChatGPT and other open models; community reception on GitHub and LinkedIn has been broadly positive about having a from-scratch Vietnamese model rather than an English model with a translation layer.

The caveats matter. PhoGPT is monolingual-focused and older — the headline 7.5B release dates to 2023, with lighter 4B models since — so on raw capability it now trails newer multilingual models and Vietnam's own larger sovereign efforts. Evaluation rested heavily on a small translated benchmark and VinAI's own human eval, there's no official hosted product, and self-hosting plus fine-tuning is required to get real value out of it.

Summary of public user & expert reviews, compiled by RECATOOLS.

Frequently asked questions

No — it is deliberately monolingual Vietnamese, pre-trained from scratch on a large Vietnamese corpus rather than adapted from an English base.
PhoGPT-7B5-Instruct, fine-tuned on 150,000 Vietnamese prompt-response pairs, for instruction following. VinAI has since released lighter PhoGPT-4B models if 7.5B is too heavy.
It is the default Vietnamese NLP baseline: the VinAIResearch/PhoGPT repo sits at 790+ GitHub stars and the model is routinely cited in Vietnamese NLP resource lists.

About this listing

Researched on
Published on
Last reviewed

This entry was compiled from publicly available data including PhoGPT's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with PhoGPT unless explicitly stated.

Data accuracy

Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.

For the latest details, please refer to PhoGPT directly →

Spotted something out of date? Suggest an update →

Advertisement