Tarjama Unveils Pronoia V2, Outshining Global AI Giants in Arabic Language AI

Arabic-speaking markets have long stood in the shadows of global technology efforts, often playing second fiddle when it comes to advanced AI developments. Now, finally, something's shifting: Tarjama, a prominent Middle Eastern tech firm, recently announced the launch of its latest language model—Pronoia V2—promising a step-change in Arabic artificial intelligence.

Pronoia V2, positioned as a next-gen large language model (LLM), appears to have nudged ahead of some big global hitters, including OpenAI’s GPT-4o, DeepSeek-V3, and Cohere. Impressively, it clocked up a 76.8% average score on standard Arabic benchmarks, compared to GPT-4o’s 58.05%, DeepSeek 70B at 53.6%, and Cohere 7B trailing at 55.3%. To put it plainly, Tarjama's latest endeavour is really setting the cat among the pigeons in Arabic-language AI tools.

One particular claim caught my eye, personally, as someone who’s spent ample time battling with sluggish software and translating bits and bobs of Arabic: Pronoia can apparently deliver results using only one GPU. What this means for businesses is drastically reduced infrastructure and running costs—a tempting prospect indeed.

CTO Iyad Ahmad points out something intriguing: "Size doesn't always mean strength," he explains. "We've proven that thoughtful architecture and precision training can outperform sheer scale." In other words, clever design beats brute force every time.

Pronoia hasn't just aced general tests either: it scored an impressive 92.4% on specialised benchmarks such as MT ARC Easy, and an equally striking 83.4% on MT Openbook QA.

It's a deliberately business-friendly platform, too. Pronoia effortlessly tackles high-quality Arabic translation, automated contract reviews, customer support in multiple languages, content creation tailored for regional audiences, and document processing at lightning speed. Al Hassan notes, "We've witnessed firsthand how businesses struggle with Arabic content management. Pronoia turns day-long processes into minute-long tasks, fundamentally changing how organisations operate in Arabic-speaking markets."

This model is offered in two configurations (7B and 14B parameters), processes text, audio, and structured data in both Arabic and English, and integrates smoothly with existing software platforms. Tarjama's easy-to-deploy solutions—including both cloud-based and on-premises capabilities—are wrapped in enterprise-grade encryption, making them attractive to security-conscious corporates. In simpler terms, it's not just some fancy academic model: it's practically built for business from the ground up with Arabic at its core.

Looking further ahead, Tarjama has ambitious goals. The upcoming Pronoia Vv32 aims for even higher benchmarks, targeting an 80% average across Arabic evaluation datasets and shooting for an exceptional COMET translation quality rating of 94 or above.

For decades, Arabic AI progress has stumbled behind English-centric innovations, generally playing catch-up. However, Tarjama’s newest offering flips that script—showing that, with thoughtful design and focussed investment, Arabic AI can more than pull its wieght on the world stage.

It seems that the days of Arabic-language models being second-tier choices might well be numbered. Arageek readers might recall plenty of times we've noted the relative scarcity of robust Arabic solutions. Now, finally, it looks like the playing field might slowly be levelling out—step by carefully placed step.

🚀 Got exciting news to share?

If you're a startup founder, VC, or PR agency with big updates—funding rounds, product launches 📢, or company milestones 🎉 — AraGeek English wants to hear from you!

✉️ Send Us Your Story 👇

https://en.arageek.com/send-us-your-story-or-a-news-tip