Saudi ALLAM 34B Ranks Second Globally on BALSAM Index, Trailing Only GPT-5.2

Riyadh | EcoPulse24

In a development reshaping the landscape of Arabic artificial intelligence, the Saudi ALLAM 34B model has achieved second place globally on the BALSAM Index leaderboard, published by the King Salman Global Academy for Arabic Language. The model surpassed several international competitors and trails only the American GPT-5.2 model.

Tariq Amin, CEO of HUMAIN, announced the accomplishment on X, emphasizing that this is more than a numerical ranking - it signals a transition for Arabic models from adaptation to direct competition at global standards.

BALSAM is a benchmarking framework for Arabic language models, utilizing approximately 1,400 datasets, over 50,000 evaluation questions, and covering 67 linguistic and cognitive tasks, including deep comprehension, paraphrasing, logical reasoning, error correction, and semantic analysis. The index is designed as an impartial evaluation tool, offering APIs for governments, companies, and researchers to test their models against a unified standard.

Securing second place demonstrates:
1. Global recognition of an Arabic model's capabilities.
2. Proof that local investment in Arabic data yields results.
3. Saudi Arabia’s transformation from a technology consumer to a developer and evaluator.

ALLAM 34B was designed as an Arabic-first model, not merely translated into Arabic, marking a significant distinction in linguistic performance. Its development began under SDAIA in 2023, with versions ranging from 7B to 70B parameters. Operational ownership shifted to HUMAIN, a Public Investment Fund subsidiary, in 2025, and the HUMAIN Chat app based on 34B launched in August 2025. The model was trained by hundreds of experts and evaluators, focusing on both standard and major dialects, and supporting smooth transitions between Arabic and English.

Strategic implications include:
1. Digital language sovereignty - addressing historic underrepresentation of Arabic in global models.
2. A national standard for evaluation - reducing reliance on external benchmarks.
3. Saudi positioning as a regional LLM hub - future iterations may compete in advanced performance sectors.

Challenges remain: the gap with the top model persists, global expansion requires multilingual context testing, and the BALSAM Index, though robust, still seeks broader international recognition.

EcoPulse24’s conclusion: ALLAM 34B’s second-place finish is not just technical news but a sign of institutional maturity for Arabic AI. The question now is not whether an Arabic model can compete, but how far the next iteration can advance.