FREEDOM251

Freedom251 › Translation Software

Marian NMT

Free plan APILinuxSelf-hostedWindows

Visit Marian NMT Alternatives Pricing

Marian NMT's own site

What Marian NMT is

Marian NMT is a free, MIT-licensed neural machine translation framework written in C++. It is designed for training and translating custom models, with a pure C++ implementation and minimal dependencies. Marian supports multi-GPU training and translation on either GPUs or CPUs. Available model architectures include deep recurrent neural networks, Transformers, and multi-source models. Training tools provide built-in SentencePiece processing, dynamic mini-batches, regularization, and transfer learning. Translation components include a decoder for CPU and GPU use, a scorer for rescoring text and n-best lists, and a WebSocket server that offers translation service. CPU builds can use Intel MKL or OpenBLAS, while GPU builds can use CUDA. The framework can be compiled statically so its resulting binary can be copied for use elsewhere. It is available through API, Linux, self-hosted, and Windows environments. Marian does not include document translation, website translation, translation memory, or glossary tools. The listed plan costs nothing, with no usage limits stated.

Who it suits

Marian NMT suits companies, organizations, and research projects building or running custom translation models. It fits teams that can work with self-hosted, Linux, Windows, or API deployments.

What it does well

  • Free, MIT-licensed framework
  • Supports CPU and GPU translation
  • Multi-GPU training support
  • Includes Transformer and RNN architectures
  • Static compilation supports portable binaries

What to know first

  • No document translation tools
  • No website translation tools
  • No translation memory or glossary tools

Plans

PlanBilled and limitsPrice
Marian NMTMIT-licensed neural machine translation framework · no usage limits statedFree

As marian-nmt.github.io lists them, read 12 Oct 2026. The maker's own page has the final word.

Getting started

  1. Open the Marian NMT website
  2. Choose Linux, Windows, API, or self-hosted deployment
  3. Build with Intel MKL or OpenBLAS for CPU, or CUDA for GPU
  4. Train or translate with a supported model architecture
  5. Expose translation through the decoder or WebSocket server

Verdict

Marian NMT focuses on model training and machine translation rather than packaged translation-management features. Its broad architecture support, CPU/GPU operation, and permissive license make it suitable for technical teams creating custom deployments.

Questions

Is Marian NMT free?

Yes. The Marian NMT plan is priced at $0, and the framework uses the MIT open-source license.

Which platforms does Marian NMT support?

Platforms listed are API, Linux, self-hosted, and Windows.

Can Marian NMT translate on GPUs and CPUs?

Yes. It supports multi-GPU training and translation on GPU or CPU.

What model architectures are supported?

Supported types include deep RNNs, Transformers, and multi-source models.

How can users get help?

The project directs questions and issue reports to its Google discussion group and GitHub issue trackers.

Facts from Marian NMT's own pages

API access
Yes · marian-nmt.github.io, 22 Sept 2026
Deployment
Marian can be compiled statically, and its features page says the resulting binary can be copied and used elsewhere. · marian-nmt.github.io, 12 Oct 2026
Document translation
No · marian-nmt.github.io, 22 Sept 2026
Glossary tools
No · marian-nmt.github.io, 22 Sept 2026
Implementation
The framework uses a pure C++ implementation and is described as having minimal dependencies. · marian-nmt.github.io, 12 Oct 2026
Integrations
The documentation describes using Intel MKL or OpenBLAS for CPU builds and CUDA for GPU builds. · marian-nmt.github.io, 12 Oct 2026
Intended users
The site describes Marian as being used by companies, organizations, and research projects, and provides tools for training and translating with custom models. · marian-nmt.github.io, 12 Oct 2026
License
Marian is released under the permissive MIT open-source license. · marian-nmt.github.io, 12 Oct 2026
Model architectures
Supported model types include deep RNNs, Transformers, and multi-source models. · marian-nmt.github.io, 12 Oct 2026
Platforms
The project documents builds for Linux and Windows; the Windows build supports both CPU and GPU. · marian-nmt.github.io, 12 Oct 2026
Purpose
Marian is a neural machine translation framework written in C++. · marian-nmt.github.io, 12 Oct 2026
Support
The project directs users to its Google discussion group and GitHub issue trackers for questions and issues. · marian-nmt.github.io, 12 Oct 2026
Training features
Training features include built-in SentencePiece processing, dynamic mini-batches, regularization, and transfer learning. · marian-nmt.github.io, 12 Oct 2026
Translation and training
Marian supports multi-GPU training and translation on GPU or CPU. · marian-nmt.github.io, 12 Oct 2026
Translation memory
No · marian-nmt.github.io, 22 Sept 2026
Translation tools
The toolkit includes a decoder for CPU and GPU translation, a scorer for rescoring text and n-best lists, and a WebSocket server that provides translation service. · marian-nmt.github.io, 12 Oct 2026
Website translation
No · marian-nmt.github.io, 22 Sept 2026

Where Marian NMT stands on Freedom251

Look at these too