Commissioned Research
Top 10 Best Nmt Software of 2026

GITNUXSOFTWARE REVIEW

General Knowledge

Top 10 Best Nmt Software of 2026

Editorial roundup of nmt software options for teams, with ModernMT and tradeoffs to help match use cases and constraints.

19 min readUpdated Editorially scoped research
How we ranked these tools
01Editorial Scope

Vendors set the topic, but our editors set the inclusion criteria and the evaluation framework before research starts.

02Source Research

Agents extract product evidence with provenance, then collect written reviews and transcribe video reviews and podcasts.

Read our full methodology →

Framework: Features 40% · Ease 30% · Value 30%

Gitnux may earn a commission through links on this page. This does not influence our reviews. Editorial policy

NMT software teams use in production to translate text and documents with neural models, typically through APIs, engine orchestration, or custom training pipelines. This comparison helps analysts and operators weigh integration depth, automation and routing controls, configuration and data model fit, and deployment options when standard machine translation does not meet domain or quality requirements.

ModernMT is the best pick for localization teams that need repeatable NMT behavior in automated pipelines because it lets terminology and translation memory be enforced during generation via API-driven configuration, while Google Cloud Translation API is a strong alternative for cloud teams running high-volume API translation queues since its async batch jobs reduce client orchestration work.

Editor’s top 3 picks

Three quick recommendations before you dive into the full comparison below — each one leads on a different dimension.

Editor pick
1

ModernMT

Terminology and translation memory can be enforced during generation through API-driven workflow configuration.

Best for: Fits when localization teams need repeatable NMT behavior with terminology and memory in automated pipelines.

2

Google Cloud Translation API

Editor pick

Asynchronous batch translation jobs reduce client orchestration work for high-volume translation queues.

Best for: Fits when cloud teams need NMT translation via API with async batch jobs and managed operations.

3

Amazon Translate

Editor pick

Terminology configuration lets domain terms override general translations during API and batch jobs.

Best for: Fits when AWS-centric teams need controlled translation automation with batch throughput and API integration.

Comparison Table

1
ModernMTBest overall
enterprise
9.0/10
Overall
2
8.9/10
Overall
3
8.8/10
Overall
4
8.3/10
Overall
5
enterprise
8.1/10
Overall
6
7.7/10
Overall
7
7.5/10
Overall
8
7.0/10
Overall
9
enterprise
6.9/10
Overall
10
enterprise
6.5/10
Overall
#1

ModernMT

enterprise

ModernMT is an open-source adaptive neural machine translation engine designed for enterprise scalability.

9.0/10
Overall
Features9.0/10
Ease of Use9.0/10
Value9.0/10
Standout feature

Terminology and translation memory can be enforced during generation through API-driven workflow configuration.

ModernMT runs NMT models for source-target language pairs and supports workflow patterns that combine batch translation throughput with terminology and memory lookups. Its integration model focuses on calling translation via API while binding business data such as terms and stored translations into the generation step. Admin controls include user access management and audit visibility for operations teams that need traceability. Automation-friendly configuration lets teams standardize outputs across projects instead of re-creating settings per request.

A tradeoff appears in setup depth, because consistent terminology and memory usage requires disciplined onboarding of termbases and quality gates for data. ModernMT fits best when teams already have translation assets and want repeatable behavior in automated pipelines rather than manual post-editing only. It is less suited to ad hoc translation needs that do not justify termbase and memory integration work.

Pros
  • +API-first integration for pipeline automation and request-level controls
  • +Terminology and translation memory integration to reduce lexical drift
  • +Audit trails and access controls for translation service governance
  • +Batch translation handling for consistent throughput in localization workflows
Cons
  • –Consistent results require structured onboarding of terms and memory
  • –Deep configuration can add coordination overhead across teams
  • –Complex workflows may need more orchestration outside the core service
Use scenarios
  • Localization engineering teams

    Automated translation pipeline with enforced terms

    More consistent terminology across releases

  • Global support operations

    Batch translation for ticket backlogs

    Lower turnaround for multilingual queues

Show 1 more scenario
  • Content governance teams

    Controlled access and traceability for NMT

    Audit-ready operational oversight

    Admin controls track translation service usage and restrict who can configure workflows.

Best for: Fits when localization teams need repeatable NMT behavior with terminology and memory in automated pipelines.

#2

Google Cloud Translation API

API-first

Google's cloud-hosted neural machine translation API providing real-time text and document translation across over 100 languages.

8.9/10
Overall
Features8.9/10
Ease of Use8.9/10
Value8.9/10
Standout feature

Asynchronous batch translation jobs reduce client orchestration work for high-volume translation queues.

Google Cloud Translation API is a good fit for teams that need neural machine translation through an application API rather than an on-premise NMT engine installation. The batch translation capability supports large payloads via asynchronous jobs, which helps reduce client-side orchestration complexity for throughput-focused work. The request model includes parameters for source language hints, target language selection, and output options for whitespace and markup behavior.

A key tradeoff is that the service model is managed by Google, so controlling fine-tuning, domain adaptation, or custom NMT training is not part of the standard API workflow. It fits situations where translation must run inside an existing cloud governance model with service accounts and audit-oriented logging, and where teams prefer operational simplicity over custom model ownership.

Pros
  • +REST API supports both synchronous and async batch translation jobs
  • +Language detection reduces preprocessing logic for mixed-language inputs
  • +Request options control formatting and output behavior for production text
  • +Integrates with Google Cloud auth, IAM enforcement, and logging workflows
Cons
  • –Custom model training and fine-tuning are not exposed in the API workflow
  • –Complex termbase and translation memory pipelines require external components
Use scenarios
  • Product localization teams

    Batch translation for release content

    Faster localization cycles

  • Customer support analytics

    On-demand translation for chat logs

    Consistent downstream metrics

Show 2 more scenarios
  • Content platform engineering

    API translation in web and mobile flows

    Lower integration complexity

    A single translation endpoint supports multiple target languages with request-level behavior controls.

  • Compliance and security teams

    Governed translation under IAM

    Stronger access control

    Service account access and centralized logging support audit-friendly operation in cloud environments.

Best for: Fits when cloud teams need NMT translation via API with async batch jobs and managed operations.

#3

Amazon Translate

API-first

Amazon Translate is a neural machine translation service for localizing content across multiple languages.

8.8/10
Overall
Features8.8/10
Ease of Use8.8/10
Value8.8/10
Standout feature

Terminology configuration lets domain terms override general translations during API and batch jobs.

Amazon Translate provides two primary execution modes, synchronous translation for request-response use cases and asynchronous batch jobs for higher throughput workloads. The API surface supports translating large documents by chunking input text externally and retrieving translated results from job outputs. For workflow integration, AWS IAM roles and fine-grained permissions control who can call translation operations and start jobs.

A common tradeoff is limited control over model internals, since the service exposes configuration around inputs and terminology rather than encoder-decoder architecture choices. Amazon Translate fits teams that need translation automation with minimal ML engineering and that can validate quality with human review loops for specific content types.

Pros
  • +Synchronous and asynchronous translation modes for mixed latency needs
  • +AWS IAM and role-based access control for governed API access
  • +Batch jobs support large text volumes without custom infrastructure
  • +Terminology support improves consistency for known vocab
Cons
  • –Limited ability to control training, model selection, or fine-tuning internals
  • –Document-level quality depends on external chunking strategy
Use scenarios
  • Customer support operations teams

    Translate inbound tickets in real time

    Faster triage with consistent terminology

  • Localization engineering teams

    Batch translate product catalogs

    Higher throughput localization cycles

Show 2 more scenarios
  • Developer platforms teams

    Embed translation into internal services

    Managed translation without model ops

    Use the translation API with IAM permissions and consistent terminology overrides.

  • Compliance and governance teams

    Restrict translation access by role

    Audit-friendly access boundaries

    Apply AWS IAM policies to control who can start and read batch job results.

Best for: Fits when AWS-centric teams need controlled translation automation with batch throughput and API integration.

#4

DeepL Translator

enterprise

DeepL translates text, documents, and supported conversations across multiple languages.

8.3/10
Overall
Features8.3/10
Ease of Use8.3/10
Value8.3/10
Standout feature

Document translation that preserves much of source formatting in the translated draft.

DeepL Translator fits NMT pipelines when translation output quality matters for external communication, internal knowledge bases, and customer-facing text. Document translation keeps much of the source structure so business users can work from a translated draft without rebuilding layout from scratch. For recurring content, terminology controls help reduce variation across language versions when marketing or support teams translate similar materials repeatedly.

Pros
  • +High-quality translations that retain tone for business communication
  • +Document translation keeps much of source structure in the draft output
  • +Terminology controls help stabilize wording across repeated content
  • +API support supports automation in translation workflows
Cons
  • –Governance controls like deep RBAC and audit logging can be limited
  • –Document translation adds upload and format handling overhead
  • –Terminology control value depends on disciplined term management
  • –Custom workflow needs may require API integration work
Use scenarios
  • Business operations teams

    Translate policies, contracts, and annexes

    Draft translations ready for review

  • Localization content teams

    Standardize recurring product and support text

    More consistent terminology across languages

Show 2 more scenarios
  • Consumers and professionals

    Translate messages and short drafts

    Quicker understanding and reply

    Web and desktop translation support fast readable translations for routine communication.

  • Teams with Chinese content

    Translate Chinese and convert variants

    Readable Chinese translations and conversion

    Workflows support Chinese document translation and Simplified-to-Traditional conversion when needed.

Best for: Fits when teams need high-quality NMT output with both document drafts and API-driven translation automation.

#5

Language Weaver

enterprise

Machine translation platform for enterprise and localization workflows with neural translation capabilities.

8.1/10
Overall
Features8.1/10
Ease of Use8.1/10
Value8.1/10
Standout feature

Job-level pipeline configuration that combines decoding controls with terminology and quality post-check steps.

Language Weaver provides an NMT pipeline surface for producing translations with configurable model behavior and translation workflows. It supports batch-oriented translation runs with controllable decoding settings and post-processing hooks for quality checks and terminology handling.

Pros
  • +API-first job orchestration for batch translation workflows
  • +Configurable decoding controls for beam search and output formatting
  • +Terminology and quality checks fit into translation job post-processing
  • +Repeatable pipeline configurations support consistent reruns
Cons
  • –Less suited to low-latency per-request translation without careful batching
  • –Advanced customization requires more setup than basic pass-through translation
  • –Workflow routing depends on external systems for review and approvals
  • –Limited visibility into token-level model behavior for deep debugging
Use scenarios
  • Localization operations teams

    Run batch translations with job settings

    More predictable output

  • Content platform teams

    Trigger translation jobs via API

    Automated translation requests

Show 1 more scenario
  • Terminology-focused translators

    Apply terminology rules in workflow

    Reduced glossary drift

    Language Weaver applies terminology handling during configured translation workflows for controlled term usage.

Best for: Fits when teams need API-driven batch NMT pipelines with terminology controls and repeatable reruns.

#6

KantanMT

SMB

Custom machine translation platform for training and deploying domain-specific neural translation engines.

7.7/10
Overall
Features7.7/10
Ease of Use7.7/10
Value7.7/10
Standout feature

Glossary-based terminology enforcement tied to API translation jobs, so consistency rules apply per request batch.

KantanMT is a neural machine translation service focused on building translation workflows around managed NMT inference. It supports API-driven translation requests for batch decoding and production routing, which helps fit KantanMT into existing pipeline code.

Pros
  • +API-first translation workflow fits into existing orchestration code
  • +Batch decoding support improves throughput for high-volume jobs
  • +Terminology and glossary controls reduce output drift across runs
  • +Project-scoped configuration keeps environment separation clean
Cons
  • –Limited visibility into model internals compared to self-hosted stacks
  • –Custom fine-tuning workflows are not the focus for iterative tuning loops
  • –Quality evaluation hooks are less granular than dedicated scoring pipelines
  • –Advanced routing requires more external glue than UI-driven tools
Use scenarios
  • Software teams needing MT APIs

    Submit batch translation requests

    Reduced custom MT plumbing

  • Localization managers

    Enforce glossary in translation jobs

    Lower variation in terms

Show 1 more scenario
  • Production operations teams

    Scope projects for controlled runs

    More consistent production output

    KantanMT uses project scoping and repeatable job patterns to manage translation behavior without training models.

Best for: Fits when teams need API-driven NMT translation runs with terminology consistency, not model training control.

#7

Intento Translator Hub

enterprise

Enterprise translation orchestration software that routes content across multiple neural machine translation engines.

7.5/10
Overall
Features7.5/10
Ease of Use7.6/10
Value7.4/10
Standout feature

Translation job orchestration via API that lets teams standardize endpoint routing and operations across projects.

Intento Translator Hub is an NMT pipeline management system that focuses on operational control around translation endpoints rather than model training UI. It supports workflow configuration for batch and request-driven translation, including routing through configurable providers and monitoring translation outcomes.

Pros
  • +API-led translation operations fit NMT pipelines with existing orchestration
  • +Configurable routing supports multi-endpoint translation workflows
  • +Project-level settings reduce drift across batch and on-demand jobs
  • +Operational monitoring supports faster triage of translation failures
Cons
  • –Fine-tuning and training pipeline tooling is not its primary workflow
  • –Advanced terminology controls require careful external termbase integration
  • –Quality evaluation hooks are limited to what the API surfaces
  • –Throughput tuning depends on upstream batching and provider limits
Use scenarios
  • Globalization engineering teams

    Route translation requests across providers

    Provider failover support

  • Translation operations managers

    Monitor outcomes across translation projects

    Better operational visibility

Show 1 more scenario
  • Developers building translation apps

    Call admin and translation APIs

    Faster integration

    Intento Hub exposes a documented API surface for project actions and translation execution without interactive tooling dependence.

Best for: Fits when teams need controlled NMT execution with API automation and multi-endpoint routing.

#8

TextUnited MT Hub

SMB

Translation management software with machine translation integration, engine routing, and quality controls.

7.0/10
Overall
Features7.0/10
Ease of Use7.0/10
Value7.0/10
Standout feature

Terminology and term consistency handling is built into the MT request pipeline so NMT outputs follow controlled lexical rules.

TextUnited MT Hub is an NMT pipeline orchestrator from TextUnited that connects translation workflows to configurable translation providers and linguistic assets. It focuses on routing and automation for batch translation, termbase and terminology controls, and post-translation handling through a structured integration surface.

Pros
  • +API-first request handling for automated translation jobs in applications
  • +Configurable terminology controls for consistency during NMT execution
  • +Workflow routing support for managing translation provider selection
  • +Batch-oriented processing options suited for throughput-focused pipelines
Cons
  • –Governed terminology setup requires ongoing maintenance to stay accurate
  • –Advanced quality estimation and human feedback loops are not exposed as a native module
Use scenarios
  • Enterprise localization teams

    Orchestrate provider workflows for batches

    Standardized batch execution

  • Terminology management teams

    Apply termbase controls across languages

    More consistent terminology

Show 1 more scenario
  • Workflow automation engineers

    Integrate post-translation handling steps

    Cleaner downstream ingestion

    The hub structures post-translation operations through an integration surface for downstream processing and governance.

Best for: Fits when translation programs need API-driven NMT orchestration plus terminology controls across multiple production systems.

#9

Fairseq

enterprise

Fairseq is an open-source neural sequence modeling toolkit for training custom machine translation models.

6.9/10
Overall
Features6.9/10
Ease of Use6.9/10
Value6.9/10
Standout feature

Fairseq’s task and generation abstractions let custom translation datasets and decoding strategies plug into the same training codebase.

Fairseq is a research-first NMT training toolkit from Meta that ships ready-to-run encoder-decoder models such as Transformer. It provides a configurable training and inference loop for tasks like translation, and it supports common decoding workflows like beam search and batch generation.

Pros
  • +Config-driven training and inference loops for repeatable translation experiments
  • +Strong PyTorch extensibility through custom modules and criterion implementations
  • +Built-in beam search decoding and standard evaluation scripts for MT metrics
  • +Checkpoint-based fine-tuning workflows that reuse prior model runs
Cons
  • –Production orchestration features like job queues and workflow scheduling are not native
  • –End-to-end data pipeline handling is thin versus dedicated NMT pipeline tools
  • –Model deployment targets are mostly DIY export and scripts, not managed serving
  • –Large-scale throughput tuning requires deeper engineering in training and batching
Use scenarios
  • Academic ML researchers

    Train Transformer-based NMT models

    Reproducible model training

  • ML engineers

    Run beam search batch generation

    Comparable decoding outputs

Show 1 more scenario
  • Applied NLP teams

    Customize NMT components in PyTorch

    More tailored model behavior

    Fairseq’s extensible PyTorch codebase enables custom architectures and training recipes within translation pipelines.

Best for: Fits when teams build custom NMT training, evaluation, and experiment-driven fine-tuning with PyTorch.

#10

Moses

enterprise

Moses is a statistical machine translation system that includes neural model components for advanced translation pipelines.

6.5/10
Overall
Features6.5/10
Ease of Use6.5/10
Value6.5/10
Standout feature

End-to-end experiment scripts for training and decoding that keep preprocessing and inference tied to the same artifact lineage.

Moses by statmt.org provides an NMT pipeline for training and running neural machine translation models with transformer-style encoder-decoder architectures. It focuses on reproducible dataset preparation, model training scripts, and batch decoding workflows built for translation experiments.

Pros
  • +Reproducible training and decoding workflows built around saved model artifacts
  • +Batch translation setup supports throughput-focused experiment runs
  • +Clear CLI-driven experiment structure reduces hidden state across runs
  • +Works well when pipelines need deterministic preprocessing and labeling
Cons
  • –Operational features like RBAC and audit log are not provided as built-in controls
  • –API-driven orchestration and fine-grained automation hooks are limited
  • –Production deployment needs external wrapping around inference and routing
  • –Tuning experiment configuration can require research-grade workflow discipline
Use scenarios
  • Machine translation researchers

    Run repeatable NMT training experiments

    Consistent experiment results

  • NLP engineering teams

    Batch decode corpora for evaluations

    Faster evaluation cycles

Show 1 more scenario
  • Applied MT teams

    Wrap training in custom pipelines

    Pipeline integration via files

    Moses emits model checkpoints and configs that external orchestration layers can schedule and manage.

Best for: Fits when research teams need controlled NMT training and batch decoding with repeatable configs.

Conclusion

After evaluating 10 nmt software, ModernMT stands out as our overall top pick — it scored highest across our combined criteria of features, ease of use, and value, which is why it sits at #1 in the rankings above.

Our Top Pick
ModernMT

Use the comparison table and detailed reviews above to validate the fit against your own requirements before committing to a tool.

Frequently Asked Questions About Nmt Software

How do modern NMT tools expose APIs for pipeline automation?
Google Cloud Translation API exposes a REST interface for batch jobs and per-request translation, which fits job orchestration and async queues. Amazon Translate and DeepL also provide API-based translation endpoints, while ModernMT and Intento Translator Hub focus on API-driven workflow configuration for repeatable translation runs.
Which platforms provide terminology controls that apply during generation?
ModernMT enforces terminology and translation memory behavior through API-driven workflow configuration. DeepL uses glossary-style constraints to keep repeated translations consistent across automated requests, and KantanMT ties glossary-based terminology enforcement to API translation jobs.
When does asynchronous batch execution matter for translation throughput?
Google Cloud Translation API supports asynchronous batch jobs, which reduces client-side orchestration work for high-volume queues. TextUnited MT Hub also routes batch requests through a managed pipeline, while Amazon Translate offers batch translation jobs designed for higher throughput patterns.
What breaks if a workflow depends on interactive UI tuning but the tool is API-first?
Fairseq is built for training and checkpoint-driven experiments, so interactive UI tuning is not the core workflow and teams must rely on configuration and code changes for decoding behavior. Intento Translator Hub and Language Weaver centralize workflow behavior in pipeline configuration and API provisioning, so teams need to encode tuning decisions into job configs rather than manual UI steps.
How should teams handle data migration when moving from file-based MT pipelines to service-based APIs?
Moses integration typically centers on file-based corpora and model artifacts, so migrating to service-based APIs means converting stored inputs into request payloads and aligning output formats. Google Cloud Translation API, Amazon Translate, and DeepL accept API requests directly, which shifts migration work toward schema mapping for inputs, batch boundaries, and text normalization rules.
Which toolkits best support model training and decoding experiments instead of production orchestration?
Moses focuses on reproducible dataset preparation, training scripts, and batch decoding tied to artifacts, which fits research pipelines. Fairseq supports training and inference loops for encoder-decoder models with task and generation abstractions, while Intento Translator Hub and TextUnited MT Hub focus on orchestrating translation endpoints rather than training models.
Where do admin controls and audit logs fit in an NMT production setup?
ModernMT includes governance features such as audit trails and access control for operating translation services at scale. Intento Translator Hub also treats project-level configuration as the control plane for admin actions, while DeepL’s API-centric approach supports repeatable configuration and logging in the client application.
How can teams plug NMT translation into existing review and publishing workflows?
TextUnited MT Hub routes MT requests through a structured integration surface and carries terminology and post-translation handling into downstream systems. Language Weaver offers API-first provisioning of translation jobs with post-processing hooks for quality checks, and Intento Translator Hub supports routing through configurable providers with monitoring tied to translation outcomes.
What tradeoff appears when moving from custom PyTorch-based toolkits to managed translation services?
Fairseq enables extensibility in PyTorch code and checkpoint-driven experimentation, so custom architecture or decoding changes are easier to implement inside the training codebase. Managed services like Amazon Translate and Google Cloud Translation API reduce operational overhead but constrain model behavior control to request-level options rather than code-level training and generation rewrites.

Tools reviewed

Primary sources checked during evaluation.

Referenced in the comparison table and product reviews above.

How to Choose the Right Nmt Software

Options differ mainly in Model quality on real text and Output controllability. We also see practical gaps in API integration depth, governance of translation assets, and how well each tool handles your language pair and volume.

Quality validation on your test set

Why it matters

Quality validation on your test set is where we separate “looks good in examples” from reliable results on your actual language mix and writing style.

How to check

Prepare a small spreadsheet of 50–200 real sentences per direction (source→target and target→source if needed). In the demo, run those exact strings and compare meaning consistency and terminology usage across outputs.

Output control for minimal post-editing

Why it matters

Output control matters because small phrasing changes can break downstream readability or terminology rules, forcing more post-editing.

How to check

Ask the demo to show how you can constrain output for tone or terminology (for example, “keep product terms unchanged”). Then try paraphrase-like prompts or near-duplicate sentences and see whether outputs stay consistent without manual fixes.

API and automation surface

Why it matters

API and automation surface determines whether the tool fits into production workflows or stays a manual translation step.

How to check

Request a live call or script walkthrough that submits text and retrieves translations, including how you pass context/metadata. Confirm whether batch requests, webhooks/status callbacks, and job tracking are available in the same interface.

Terminology assets and governance

Why it matters

Terminology assets and governance reduce drift across documents and teams, which is critical for repeated terms and style.

How to check

In the demo, import a small glossary and show whether it applies per project, per user role, or per workflow step. Check whether you can audit changes to the glossary and see which version was used.

Language pair coverage and direction control

Why it matters

Language pair coverage and direction control affect whether the tool handles your required pairs correctly and consistently, especially when you reuse outputs in the opposite direction.

How to check

If you need multiple pairs, test each direction with the same content set. Ask whether the tool lets you select direction explicitly and whether it changes output behavior when you flip source/target.

Red flags

  • Outputs match demo samples but degrade on your repeated or domain-specific sentences.
  • You can’t reproduce terminology rules consistently across similar inputs.
  • The workflow requires manual steps because the API lacks job tracking.

Fit by situation

  • If your team has an existing set of real strings, prioritize Quality validation on your test set.
  • If translations must run inside production workflows, prioritize API and automation surface.
  • If you rely on strict recurring terms, prioritize Terminology assets and governance.
Commissioned Research

Funding and competing interests

This study was co-funded by several of the vendors assessed in it. No single vendor funded it. Funders could propose the market studied.

They had no role in selecting vendors, setting or weighting criteria, analysing data or drawing conclusions, and they received the same factual-accuracy check as every other vendor assessed.

Market definition, inclusion rules and criteria weights were fixed by an editor before research began and did not change afterwards. Editorial process →

Keep exploring

FOR SOFTWARE VENDORS

Not on this list? Let’s fix that.

Our best-of pages are how many teams discover and compare tools in this space. If you think your product belongs in this lineup, we’d like to hear from you—we’ll walk you through fit and what an editorial entry looks like.

Apply for a Listing

WHAT THIS INCLUDES

  • Where buyers compare

    Readers come to these pages to shortlist software—your product shows up in that moment, not in a random sidebar.

  • Editorial write-up

    We describe your product in our own words and check the facts before anything goes live.

  • On-page brand presence

    You appear in the roundup the same way as other tools we cover: name, positioning, and a clear next step for readers who want to learn more.

  • Kept up to date

    We refresh lists on a regular rhythm so the category page stays useful as products and pricing change.