Driving equitable AI access for underrepresented languages with the 3C3H Leaderboards Initiative

The rapid rise of large language models (LLMs) is transforming industries, improving efficiency, driving innovation, and unlocking unprecedented opportunities for organizations and industries worldwide. In the United States, 45% of organizations are leveraging LLMs for decision-making and automation. This momentum is accelerating, with the global LLM market projected to reach $224 billion by 2034, fueled largely by content generation and enterprise applications. 

Yet, these advancements remain concentrated in high-resource languages such as English and Mandarin, leaving nearly 7,000 underrepresented languages at threat of being left behind. This linguistic divide limits AI accessibility, reinforcing inequities in technology adoption and digital transformation.

As AI continues to develop and reveal new opportunities for industries and economies globally, it’s become critical to ensure its benefits are being extended to the communities that need them most.

Introducing the 3C3H Leaderboards Initiative

At Inception, we are committed to making AI linguistically inclusive, culturally aware, and globally accessible. In 2023, we redefined AI accessibility with the launch of JAIS, the world’s most advanced Arabic Large Language Model. In 2024, we expanded our impact with NANDA, a cutting-edge Hindi Large Language Model. And today, we are taking a massive leap in our mission with the launch of the 3C3H Leaderboards initiative, a pioneering effort to establish new standards for evaluating LLMs in underrepresented languages.  

The 3C3H Leaderboards initiative creates a decentralized evaluation network by partnering with global communities, research labs, and universities to create high-quality, native and culturally aware generative benchmarks, ensuring AI models serve diverse linguistic populations. Through computing sponsorships and technical support, we empower these communities to build public leaderboards. In return, they contribute to culturally aware generative benchmarks on a periodic basis (every three to six months depending on the language), driving ongoing and sustainable support.

A major milestone of this initiative is the first-ever Moroccan Arabic generative leaderboard, developed in collaboration with AtlasIA and MBZUAI France Lab, marking a significant step toward linguistic inclusivity in AI. Beyond Moroccan Arabic, and as part of this initiative, we are developing benchmarks for Hindi, Kazakh, Southeast Asian and African languages, with contributions from internal members and independent experts worldwide.

Redefining AI Benchmarking: The 3C3H Score

The 3C3H score, first introduced as part of the AraGen Leaderboard, represents a paradigm shift in AI benchmarking. Unlike conventional static, multiple-choice evaluations, it deploys dynamic, generative, and multi-turn task-based assessments to measure AI performance across six critical dimensions: correctness, completeness, conciseness, helpfulness, honesty, and harmlessness. 

With the 3C3H score, we can robustly and holistically measure LLM performance across any language, along with identifying their shortcomings to drive targeted improvements in underrepresented languages. This marks a significant step toward a more inclusive digital future, aligning with Inception’s mission to democratize AI for all.

A Global Imperative for AI Equity

The 3C3H Leaderboards initiative is a call for a fundamental shift in AI evaluation to ensure linguistic diversity and cultural alignment. By setting new evaluation standards, we are bridging the gap between technology and culture, ensuring that every language, culture, and community has a voice in AI.

Join us

Help us build a more inclusive AI ecosystem by contributing to generative benchmarks while receiving the compute sponsorship needed. Learn more at through our Hugging Face page or contact us.

Spam-free subscription, we guarantee. This is just a friendly ping when new content is out.

Warning
Warning
Warning.