QContact is proud to announce CAI Lingo, our new language detection AI supporting 69 languages, including every official EU language and all 11 official spoken languages of South Africa. In our benchmarks, CAI Lingo outperformed OpenAI’s GPT-6 Luna, Google’s CLD3, and Meta’s fastText while delivering results in well under a millisecond.
Language detection is a critical component in routing customer interactions to the right person or AI. From the first message a customer sends, understanding which language they’re using so you can both reply and route correctly means faster resolution and a better customer experience.
For multilingual markets such as South Africa and Europe, this is particularly challenging, as many traditional language detection models struggle on the types of messages that make up chat conversations – short phrases and plenty of mixed vernacular traditionally have been weak spots in traditional models.
QContact today has a solution for that – CAI Lingo – a lightweight, extremely fast language identification model designed for real-world multilingual communications.
69 languages – billions of speakers
Our model covers 69 languages including
- All 24 official European Union languages – from English, Spanish, French, Portuguese to Maltese, Irish and Estonian.
- All 11-official spoken South African languages including isiZulu, isiXhosa, isiNdebele, siSwati, Sepedi, Sesotho, Setswana, Xitsonga, Tshivenda, Afrikaans and English.
- 34 additional languages including Arabic, Mandarin Chinese, Japanese, Korean, Hindi, Swahili, Ukrainian, Turkish and more.
QContact brings accurate detection across every language to ensure seamless routing to the right person or flow with the highest level of precision.
How does our model compare?
We compared our brand-new proprietary engine against not just the leading language detection models, but also against the very latest in LLM technology.
We took a sample of all 24 official European languages and all 11 official spoken South African languages (34 languages in total), and passed each model the exact same underlying text inputs and we measure how they compare
Language detection accuracy by model (%)
- Google Gemini 3.5 Flash-Lite96.59%
- QContact Lingo95.40%
- OpenAI GPT-6 Luna94.07%
- Lingua82.94%
- Desert Ant Tongue76.56%
- Google CLD375.37%
- Meta fastText65.43%
Our model achieved 95.4% accuracy, outperforming OpenAI’s GPT-6 Luna’s 94.07% on and coming within 1.19 percentage points of Google’s Gemini 3.5 and outperforming every other language detection model by significant margin.
For South African customers the results are even more stark with most vendors not covering all 11 languages, and with much lower levels of accuracy, QContact beats every other model including state-of-the-art LLM models.
South African language detection accuracy by model (%) · 11 official languages
- QContact LingoAll 11 languages92.52%
- Google Gemini 3.5 Flash-LiteAll 11 languages89.72%
- OpenAI GPT-6 LunaAll 11 languages83.18%
- Lingua7 languages61.21%
- Google CLD35 languages41.59%
- Tongue4 languages35.51%
- Meta fastText2 languages14.02%
And beyond our core 34 languages, a broader 54-language diagnostic dataset, our model achieved 97.01% accuracy, exceeding Tongue’s 96.40%, Lingua’s 93.67%, CLD3’s 87.67% and fastText’s 86.21%.
From seconds to fractions of a millisecond with full data sovereignty
Accuracy though is only part of the story.
Our clients handle millions of messages every day making speed just as important as accuracy. CAI Lingo recorded an average execution time of just 0.098ms. That’s over 8,000 times faster than the median response time recorded for OpenAI and Gemini APIs.
Better still, CAI Lingo runs locally on our infrastructure within each region, keeping language detection fast, cost-efficient and without transmitting customer data to third-party AI providers or leaving the country.
And we’re not finished
CAI Lingo is our second in-house model, following the launch of our advanced speech recognition engine earlier this year. Together, these technologies represent our continued investment in building a true multilingual platform that understands customers in the languages they actually speak.
And the best part? CAI Lingo is available to every QContact customer today at no additional cost.
So, if you are looking for a vendor who can deliver solutions for your customers, why not speak to QContact today.
Benchmark methodology: QContact conducted matched evaluations using identical text inputs across participating models. Accuracy represents correct language classifications, with incorrect and unknown responses counted against the total. Model coverage varies, and some baselines do not support every evaluated language. Local inference timings and external API response times use different measurement boundaries. Results may vary by language, input length and dataset.
