Riyadh-based AI professional Hesham Haroon has launched the Arabic AI Atlas, an open-source map and repository designed to help developers find Arabic-capable models, datasets and tools in one place.
The free resource lists more than 1,000 models and tools, along with more than 1,000 datasets covering large language models, speech recognition, text-to-speech, optical character recognition and embeddings. Each entry includes details such as country, organisation, licence, size, Hugging Face downloads and last update.
Users can browse an interactive country map, filter entries by dialect or licence, and share filtered views. The repository also works as a Claude Code plugin, with an offline Model Context Protocol server that can recommend models from the Atlas data.
Saudi Arabia has the largest number of listed entries at 355, followed by the UAE with 184, Egypt with 175, Qatar with 128 and Morocco with 71. By dialect, Modern Standard Arabic leads with 441 entries, followed by mixed Arabic with 310, Maghrebi with 307 and Gulf Arabic with 170.
The Atlas identifies 15 gaps in Arabic AI development, including open Kuwaiti or Qatari text-to-speech, a Libyan Arabic corpus, Palestinian speech recognition and an open Arabic handwriting benchmark. Its code is MIT licensed, while the data is available under CC BY 4.0.
Source: Middle East AI News
