Key takeaways
- RWS and Cohere collaborate to build a specialized translation model to power the new Language Weaver Pro
- The model leverages RWS’ global language and cultural expertise and the same foundation as Cohere’s LLMs
- Comprehensive benchmarking tests show that the new model outperforms competitors in 31 out of 32 languages
In a global organization, multilingual communication is critical to success. The stakes are often high, making accuracy, security, and compliance non-negotiable. Most translation tools rely on general-purpose models that attempt to do a specialist job, resulting in inconsistent output, data exposure risk, and no accountability for inaccuracies.
Now, Language Weaver Pro, a next-generation AI translation solution from RWS, one of the largest translation service providers in the world, introduces a new standard for enterprise AI translation. The solution combines Cohere’s cutting-edge large language model (LLM) with the security, control, and domain expertise that global organizations demand.
More than just a partnership
According to RWS CEO Ben Faes, “The word ‘partnership’ understates the relationship that we’ve built with Cohere.” It’s really been a collaboration from the very beginning, when in July of last year, Cohere asked RWS to help test Command A Translate prior to launch. After working closely with the Cohere team to understand and improve the model, the RWS research team quickly realized that Cohere models offered something unique and special.
By September, the two teams had begun working on a new, specialized translation model that leveraged RWS’ global language and cultural expertise and used the same foundation as Cohere’s LLMs for use in RWS products. “It brought a fascinating energy to our product team to work so closely with a frontier model team like Cohere,” recalls Faes.
The model as the “brain” behind the product
The machine translation model that emerged from the collaboration between RWS and Cohere lives at the heart of Language Weaver Pro. According to Faes, the model is “the brain behind the product,” enabling Language Weaver Pro to deliver the highest level of AI translation quality and performance for complex and regulated environments. Comprehensive benchmarking tests have shown that the new model was more reliable than other models, including the market standard model DeepL, outperforming competitors in 31 out of 32 languages that are commonly used across enterprise domains.
Comprehensive benchmarking tests have shown that the new model outperforms competitors in 31 out of 32 languages that are commonly used across enterprise domains.
Beyond raw translation capability, the custom Cohere model offers cultural intelligence features that handle global variations in language use, making it particularly effective for diverse customer needs. For RWS, the new solution represents a shift from pure translation to language intelligence, one that can deliver meaning, fluency, and accuracy at scale. “Most AI translation speaks the language but misses the meaning,” explains Faes. “Language Weaver Pro closes that gap — it's the first AI translation solution built to understand not just words, but culture, context, and compliance.”
Built-in security, from data to deployment
Security, privacy, and compliance considerations were fundamental to the design and architecture of the Language Weaver Pro solution. The joint development team established an end-to-end secure communications loop where RWS maintains complete ownership over data flowing through the application. Cohere’s model can be deployed on only two GPUs and operates on secure RWS infrastructure.
Benchmarking the new model
Our custom model for Language Weaver Pro underwent rigorous evaluation against leading commercial translation systems. Human evaluators compared Language Weaver Pro to DeepL NextGen across sentence and paragraph translations in IT/electronics and high-end marketing domains. Automated benchmarks tested against DeepL NextGen, GPT-5, Gemini 3, and Claude Sonnet 4.5 across 32 languages. The evaluation methodology included professional translators scoring translations on a 1-5 Likert scale with randomized display order to prevent bias.
The results were impressive. Language Weaver Pro achieved 55% overall sentence test wins against DeepL NextGen in human evaluations, with consistent performance across both domains. The model maintained a 100% win rate in English-to-Japanese paragraph translations, where the test wins extended to 62%, and won three out of four tests for English-to-Czech. Faes says, “The test results have been phenomenal. I’m amazed that we’re getting to that level of quality in such a short time.”
Automated benchmarks showed Language Weaver Pro outperforming all competitors in 31 out of 32 languages. The model demonstrated particular strength at paragraph-level translations, suggesting superior contextual understanding. Performance remained robust across both factual and creative content domains, indicating domain-agnostic quality advantages.
To learn more about the Language Weaver Pro testing and benchmarking process, please read the RWS benchmarking whitepaper.
Real-world applications of language intelligence
Among the myriad of enterprise translation use cases, the model particularly shines with those that involve complex languages or specialized terminology. For example, current applications include translating difficult names and specialized formats in languages like Japanese and Chinese, where standard LLMs fail to provide accurate results. Industry-specific use cases involve technical translations of things like chemical formulas and specialized terminology that require contextual nuance.
For some contexts, such as legal and finance, human translators remain essential for complex translations. As part of the custom model project, the RWS and Cohere teams are working to enhance their efficiency, potentially reducing the number of human reviewers needed post-LLM translation while maintaining quality standards, thus allowing RWS more bandwidth to grow the business. This could also significantly benefit translation of difficult-to-access languages like Icelandic and other native languages where qualified translators are scarce, as well as address current gaps in the translation ecosystem, particularly for specialized content requiring human-level understanding.
The team is also exploring other applications for Language Weaver Pro, such as adding image and voice translation capabilities, as well as optimizing the model for running in an edge environment or on limited hardware.
Collaborative partnerships: A winning approach
Strategic AI partnerships can create competitive differentiation in specialized markets. Our partnership with RWS shows how combining domain-specific expertise with cutting-edge AI capabilities can deliver measurable business value, in this case, superior translation quality that directly impacts customer satisfaction and operational efficiency.
The launch of Language Weaver Pro also highlights the growing importance of security-first AI solutions for enterprise deployments, particularly in regulated industries where data governance and compliance are paramount concerns.
CEO Ben Faes sums it up: “In the enterprise space, the partnership is what really matters. We’re on this journey together, which is a differentiator for us.”
To learn more about the power of customizing Cohere models, please reach out to us.