April 2026 Product Update

Translation has never really been about just moving words from source to target.

It's always been about thoughtful silent interactions with text. Looking in between the words, squinting at every clue of context and subtext, understanding target audiences. The best translators have always been masters at these skills.

But now we are entering a new era, where these skills can be modelled and deployed at an unprecedented level, making the translation process more sustainable and scalable.

We are preparing wxrks as a true agentic workforce, and Model Routing is a cornerstone of this new infrastructure, giving you the power to assign the exact right intelligence to the right task.

But the most important shift isn't the models, it's how you interact with them.

Human insight remains the core of everything we do. However, moving forward, that insight won't be spent on repetitive manual fixes. Your focus will progressively shift to architectural setup, strategic tuning, and orchestration.

We are building wxrks to capture every edit, every adjustment, and every management action your users take. Our goal is to squeeze every last drop of intelligence from your expertise and feed it back into your engine.

This is the foundation of a self-tuning, iterative learning machine.

You architect the system. The system scales your insight.

And without further ado...

Say good-bye to the "best" model

There is no "best" model.

There is only the best model for the specific task.

Today, we are launching Model Routing.

We built a flexible framework that allows you to configure specific models for specific agents in specific places. This isn't just a feature, it’s a new paradigm for AI.

Precision Architectures

You no longer have to settle for a single generalist. You can now build a specialized AI workforce tailored to your content type:

• Legal Workflows: Use Claude Opus 4.6 for Context Sensitivity and Gemini 3 to power Smells.
• Marketing Workflows: Swap in GPT-5.4 for Context Sensitivity and Llama 4 for Semantic Evaluation.

Models are differentiating rapidly in reasoning, speed, and cost. Some are 30x faster, others are 20x cheaper. Model Routing lets you exploit those differences.

The Power of Perspective

By getting different models to talk to each other, you unlock a richness of perspective. When models collaborate, they validate, challenge, and refine outcomes in a way no single LLM can.

The Agentic Translation Workforce

Every agent still runs through the proprietary orchestration, data-processing, and validation layers we’ve built. This update is a foundational step in transforming wxrks into an agentic workforce that scales, whether you are an individual wxrks user or a global enterprise.

Configure How Context Is Evaluated

Layer upon layer, part of our mission is making translation technology more transparent.

There’s this concept in localization called a 100% match, which means the segment you are translating is exactly the same as a segment translated in the past and stored in Translation Memory.

Then there’s the 101% match, which means both the segment and its context are the same.

But then comes the real question: what exactly is “context”?

Is it the previous and next segments? A Context Key? Metadata? Document structure? All of them?

With these new TM Context settings, you can now specify exactly what should count as context.

No ghost in the machine decides what a 101% match means for you anymore.

I never got 101% in the first place, but assuming 101% makes sense, this feature should really be marketed as 102% match. 😂 Sorry, couldn’t resist.

When the Source Needs Love and Attention Too

Is there anything more sacred than the source?

In translation, the source text is often treated almost like a divine institution, fixed, untouchable, absolute.

But anyone who has worked with real-world content knows the truth: sometimes the problem starts in the source itself.

A sentence is unclear. A headline is too long. A paragraph was written for one audience but sent to another. And translators are left trying to preserve meaning that was never fully clear to begin with.

That’s why we built Edit Source.

Instead of treating the source as untouchable, translators can now refine, clarify, and improve the original content before translation begins. Shorten a sentence. Fix ambiguity. Adjust tone. Make the content easier for humans (and AI) to understand.

Because the future of translation isn’t simply about protecting the source at all costs.

It’s about protecting understanding.

Other Updates

🗂️ Bulk Uploads for Translation Memories

📊 Project Automation and Cost Calculation Improvements

🔥 Video Localization with Embedded Subtitles

What's New

From Berlin to San Francisco

Spring has been busy.

Over the last few weeks we shook up both GALA in Berlin and GlobaSaké in San Francisco.

At GALA, we opened for the keynote and shared something that has been central to how wxrks is built: humans are still at the center of the system.

Automation is getting easier every day. Models are improving fast. The real challenge is no longer “can you automate?”. It’s how you capture human decisions, edits, and expertise, and continuously feed that back into the engine in a meaningful way.

That’s also why we’ve been happy to see wxrks consistently demonstrating robustness across model changes and evolutions. If you haven’t seen our comparison between LLMs and wxrks orchestration, take a look, it tells an important story about why layers, context, and human feedback still matter.

How Coca-Cola Europacific Partners Is Changing Translation

30 million words translated.
22,000+ requests created.
80% user adoption in the first three months.
75% less admin overhead.

These were some of the numbers Elitza Dublewa-Servatius shared during her session at GALA WorldReady in Berlin. Elitza shared the story behind Coca-Cola Europacific Partners’ translation portal built together with wxrks. What started as a simple pain point with employees jumping between random AI tools and disconnected workflows, turned into a much bigger idea: creating one place where people across 31 markets could communicate without losing tone, context, or personality along the way.

What people really connected with was her approach. Elitza didn’t talk about AI as magic or replacement. She talked about communication. About making systems that actually work for humans. About building technology that listens instead of forcing people to adapt to it.

The session became less about translation tech and more about something bigger. How companies can use AI without losing their voice in the process.

Grateful to have you with us. Let’s move forward with purpose, together.

Let’s make it happen,
let’s make it wxrks!