As global companies increasingly rely on multilingual virtual meetings, the quality of AI translation has become a mission-critical factor for collaboration tools. A new benchmark study suggests DeepL may currently hold the edge.

The AI language technology firm revealed results from an independent evaluation conducted by research and language intelligence company Slator, comparing real-time translation and caption performance across major collaboration platforms.

The study found DeepL Voice outperformed built-in caption translation systems in Google Meet, Microsoft Teams, and Zoom, delivering stronger translation quality and more stable live captions during meetings.

For enterprises increasingly conducting cross-border negotiations, strategy sessions, and customer calls in multiple languages, those improvements could have real-world business implications.

AI Translation Is Becoming Core Infrastructure

As international teams collaborate more frequently through video conferencing platforms, real-time translation has shifted from a convenience feature to a foundational communication layer.

Even small translation errors—or unstable captions that constantly rewrite themselves on screen—can slow meetings, cause confusion, or derail discussions in high-stakes scenarios.

“Language AI is becoming the core infrastructure for how global businesses operate,” said Jarek Kutylowski, CEO of DeepL.

“In that context, accuracy and stability aren’t features—they’re requirements.”

The benchmark suggests DeepL’s voice translation system currently performs better than native caption tools included in mainstream meeting platforms.

Translation Quality: DeepL Leads the Pack

The Slator evaluation assessed translation quality using blind reviews conducted by 28 professional linguists across 14 language combinations.

Seven translations were evaluated into English, and seven from English, covering a wide range of multilingual meeting scenarios.

The results showed DeepL achieving significantly higher quality scores:

• 96.4 / 100 for DeepL Voice in Zoom
• 96.3 / 100 for DeepL Voice in Microsoft Teams

By comparison, competing caption translation systems scored between 87 and 89 across the tested platforms.

In practical terms, DeepL Voice also generated fully accurate translated segments 79% of the time, nearly double the 42% success rate seen in other tools evaluated.

Another key metric involved critical translation errors—mistakes that could meaningfully alter meaning during a conversation.

DeepL Voice reduced major or critical errors by 76% on average compared with the competing systems.

Caption Stability: The Often Overlooked Factor

While translation accuracy is the most visible benchmark, Slator’s research highlights another critical aspect of real-time communication: caption stability.

In many AI captioning systems, subtitles constantly update as speech recognition models refine their output. That behavior can cause words or phrases to flicker or rewrite themselves repeatedly.

For meeting participants reading captions in real time, that instability can disrupt comprehension.

To measure the issue, Slator conducted frame-level analysis of captions as they appeared on screen, evaluating flicker, oscillation, and rewrite frequency.

Here again, DeepL’s system performed strongly.

The platform recorded caption stability scores of:

• 88.6 / 100 for DeepL Voice in Zoom
• 85.8 / 100 for DeepL Voice in Teams

DeepL Voice also reduced caption churn—the constant rewriting of subtitles—by:

• 37.6% compared with Microsoft Teams
• 54.7% compared with Zoom

For meeting participants relying on captions to follow multilingual conversations, that stability could make discussions easier to track in real time.

Linguists Overwhelmingly Preferred DeepL

Beyond numerical scoring, the study also gathered subjective evaluations from professional linguists.

Across all blind comparisons, 96% of linguists preferred DeepL Voice over competing caption translation tools.

According to Alex Edwards, the evaluation focused not just on accuracy but on how translations behave during real-world reading.

“We didn’t just want to know if the words were right at the end,” Edwards said. “We wanted to see how captions behave while someone is trying to read them.”

That meant evaluating readability, linguistic fluency, and visual stability simultaneously.

Why Caption Behavior Matters in Meetings

Slator’s findings suggest that caption stability may be as important as translation accuracy in real-time settings.

Even when translations are technically correct, frequent updates and rewrites can break a reader’s concentration.

That disruption is especially problematic during:

• Customer negotiations
• Sales presentations
• Executive strategy discussions
• Cross-border team collaboration

In these scenarios, participants often rely on captions to keep pace with conversations happening in unfamiliar languages.

When subtitles constantly shift or flicker, comprehension can suffer—even if the final translation is correct.

How the Benchmark Was Conducted

The evaluation used a combination of human and automated testing methods.

Key elements of the methodology included:

• 28 professional linguists conducting blind evaluations
• 14 language pair combinations
• Analysis of real-time caption rendering behavior
• Comparison of standard out-of-the-box platform settings

The report compared native translation features in Google Meet, Microsoft Teams, and Zoom with DeepL Voice integrations for Teams and Zoom.

Slator emphasized that it maintained full editorial control over the evaluation process and findings, despite the study being commissioned by DeepL.

A Competitive Market for Language AI

DeepL has built a reputation in recent years for high-quality machine translation, often competing with platforms from major tech companies.

Real-time meeting translation is becoming one of the most competitive segments of the AI productivity market, as businesses increasingly expect language barriers to disappear inside digital collaboration tools.

For enterprise teams operating across continents, reliable AI translation can mean faster decision-making and smoother collaboration.

What’s Next for DeepL Voice

The benchmark results arrive ahead of a broader product update planned by DeepL.

The company says it will unveil major upgrades to DeepL Voice on April 16, 2026, including expanded capabilities for translation automation and cross-platform collaboration.

If adoption continues to grow, tools like DeepL Voice could become a standard layer across enterprise meeting platforms—quietly translating conversations in real time while teams focus on the discussion itself.

For global organizations, that could make language barriers increasingly invisible.

Get in touch with our MarTech Experts.