AceGPT v2

AceGPT v2 is the follow-up to AceGPT, the Arabic-localized large language model from a collaboration between KAUST (Saudi Arabia), the Chinese University of Hong Kong, Shenzhen, and the Shenzhen Research Institute of Big Data. The original AceGPT was described as the first open-source Arabic LLM to cover the entire pipeline: pretraining, supervised fine-tuning, and reinforcement learning from AI feedback.

The v2 release continues the localization-first recipe: further pretraining a base model on a large Arabic corpus, instruction tuning on naturally written Arabic questions rather than translated English data, and alignment with a reward model that respects local culture and values. The project also introduced ACVA, a benchmark for Arabic cultural and value alignment.

AceGPT v2 was added to the Stanford HELM evaluation framework in October 2025 as one of its Arabic language models, confirming it as a distinct maintained release. It is an open model; exact license terms should be checked on its release repository before commercial use.