Models

World model from IFM at MBZUAI simulating text-described actions as video up to 20 seconds. 29B perceptual backbone; public weights and training/inference code under Apache-2.0, with third-party licences retained. Arabic prompt support is not documented. Self-hosting required; not offered in Souq Playground or for purchase. Long-term world consistency remains unresolved. News announcement: 11 November 2026, not a confirmed first-release date.

Arabic–English reasoning model from the Applied Innovation Center with thinking and direct-answer modes, tool calling through vLLM and Apache 2.0 weights.

Six open Arabic models from 0.9B to 375B parameters, covering MSA and regional dialects.

7B language model built on Falcon-H1-Arabic for Emirati vocabulary, idioms and cultural context. TII announced it on 6 October 2026 and identified Falcon Chat as the access platform.

Lightweight Arabic OCR model extracting text, tables, mathematical formulas and document sections from images. Launched on 6 October 2026; distinct from the general Falcon-OCR repository.

428B mixture-of-experts Arabic model built on MiniMax-M3 and further trained on 1T+ Arabic tokens. Weights promised under the MiniMax licence.

40B depth-extended Arabic model built on Qwen3-30B-A3B, 20K context.

3B, 7B and 34B hybrid models leading the Open Arabic LLM Leaderboard, with 256K context.

Open Arabic LLMs at 8B and 70B, trained on a large Arabic-first dataset.

Sentence-similarity embedding model based on BAAI/bge-m3, with spoken Emirati Arabic examples in its model card. Broad Gulf, code-switching and Arabizi performance claims are not verified.

Egyptian Arabic models (4B, 12B) that read Arabic script and Arabizi.

Arabic sentence-embedding and retrieval models for semantic search and RAG.

Arabic-focused 7B model built on Falcon 3 with a 32K-token context window, covering Modern Standard Arabic, Gulf and Levantine dialects. Launched on 21 May 2025; distinct from Falcon-H1-Arabic.

Arabic-literate AI tool for media, announced at Dubai AI Week 2025.

Small enterprise-focused Arabic–English model; runs locally through Ollama.

Regional model trained for Arabic and South Asian languages and cultural context.

Arabic ModernBERT encoder with 8K context for search and classification.

7B Arabic–English model that reads images and documents (OCR, medical, satellite); usable in WhatsApp and Telegram.

Arabic-centric multimodal platform: Fanar Star and Prime LLMs plus speech and image models. Fanar 2.0 launched Dec 2025.

Arabic sentence-embedding model for semantic search.

Arabic-centric embedding models encode text for semantic and cross-dialect retrieval tasks.

Reranker that improves Arabic search and RAG results.

Language models adapted for Moroccan Darija.

Open Arabic language and voice models, plus public Arabic benchmarks.

An Arabic-focused language model family provides pretrained and chat-tuned checkpoints.

100B-parameter Arabic model on Huawei Cloud, launched in Cairo; firms can build their own local models on it.

7B Arabic–English model trained on Saudi data sets, from Riyadh-based Watad.

Small Arabic-only GPT models (0.1B–0.8B) for lightweight text generation.

An instruction-following language model supports Arabic within a set of 101 languages.

BERT models for Moroccan Darija, including Arabizi text written in Latin letters.

Saudi national model family; 7B on Hugging Face, larger sizes via watsonx and Azure. Amazon Bedrock and Microsoft Foundry announced at LEAP 2026.

Arabic LLM trained on 2T+ curated Arabic tokens, with a dialect-learning technique; also a document-OCR model.

Arabic text-to-text transformer models support generation and language-understanding tasks.

BERT models for MSA, dialects and classical Arabic, with ready-made sentiment, entity and dialect-ID versions.

Arabic transformer encoders support language understanding in formal Arabic and dialectal text.

The most-used Arabic BERT family, with over 500k downloads a month; a standard base for Arabic classification and search.

This exhibitor offers high-level computational systems designed to simulate human thought processes for complex problem-solving in enterprise settings.

This company develops specialized Liquid Foundation Models including visual, audio, and large language models that support the Arabic language.

Advanced software solutions that automatically convert structured data into natural-sounding human language.