Large language models (LLMs) developed, deployed, and researched across the GCC — including Falcon (TII), Jais (MBZUAI/Inception), AceGPT (KAUST), ALLaM (SDAIA), and Fanar (QCRI).
This paper introduces Absher, a new benchmark for evaluating LLMs' linguistic and cultural competence in Saudi dialects. The benchmark comprises over 18,000 multiple-choice questions spanning six categories, using dialectal words, phrases, and proverbs from various regions of Saudi Arabia. Evaluation of state-of-the-art LLMs reveals performance gaps, especially in cultural inference and contextual understanding, highlighting the need for dialect-aware training.
MBZUAI researchers have introduced MIRA, a novel framework for improving the factual accuracy of multimodal large language models in medical applications. MIRA uses calibrated retrieval to manage factual risk and integrates image embeddings with a medical knowledge base for efficient reasoning. Evaluated on medical VQA and report generation benchmarks, MIRA achieves state-of-the-art results, with code available on GitHub.
The authors introduce Nile-Chat, a collection of LLMs (4B, 3x4B-A6B, and 12B) specifically for the Egyptian dialect, capable of understanding and generating text in both Arabic and Latin scripts. A novel language adaptation approach using the Branch-Train-MiX strategy is used to merge script-specialized experts into a single MoE model. Nile-Chat models outperform multilingual and Arabic LLMs like LLaMa, Jais, and ALLaM on newly introduced Egyptian benchmarks, with the 12B model achieving a 14.4% performance gain over Qwen2.5-14B-Instruct on Latin-script benchmarks; all resources are publicly available. Why it matters: This work addresses the overlooked aspect of adapting LLMs to dual-script languages, providing a methodology for creating more inclusive and representative language models in the Arabic-speaking world.
Researchers from MBZUAI introduce Forget-MI, a machine unlearning method tailored for multimodal medical data, enhancing privacy by removing specific patient data from AI models. Forget-MI utilizes loss functions and perturbation techniques to unlearn both unimodal and joint data representations. The method demonstrates superior performance in reducing Membership Inference Attacks and improving data removal compared to existing techniques, while preserving overall model performance and enabling data forgetting.
This paper introduces MOTOR, a multimodal retrieval and re-ranking approach for medical visual question answering (MedVQA) that uses grounded captions and optimal transport to capture relationships between queries and retrieved context, leveraging both textual and visual information. MOTOR identifies clinically relevant contexts to augment VLM input, achieving higher accuracy on MedVQA datasets. Empirical analysis shows MOTOR outperforms state-of-the-art methods by an average of 6.45%.
A new methodology emulating fact-checker criteria assesses news outlet factuality and bias using LLMs. The approach uses prompts based on fact-checking criteria to elicit and aggregate LLM responses for predictions. Experiments demonstrate improvements over baselines, with error analysis on media popularity and region, and a released dataset/code at https://github.com/mbzuai-nlp/llm-media-profiling.
This paper benchmarks reasoning-focused LLMs, especially DeepSeek models, on fifteen Arabic NLP tasks. The study uses zero-shot, few-shot, and fine-tuning strategies. Key findings include that three in-context examples improve F1 scores by over 13 points on classification tasks, DeepSeek outperforms GPT-4-mini by 12 F1 points on complex inference tasks in the zero-shot setting, and LoRA fine-tuning yields up to an additional 8 points in F1 and BLEU. Why it matters: The systematic evaluation provides insights into the performance of LLMs on Arabic NLP, highlighting the effectiveness of different strategies for improving performance and contributing to the development of more capable Arabic language models.
A new benchmark, ViMUL-Bench, is introduced to evaluate video LLMs across 14 languages, including Arabic, with a focus on cultural inclusivity. The benchmark includes 8k manually verified samples across 15 categories and varying video durations. A multilingual video LLM, ViMUL, is also presented, along with a training set of 1.2 million samples, with both to be publicly released.
MBZUAI researchers introduce TerraFM, a scalable self-supervised learning model for Earth observation that uses Sentinel-1 and Sentinel-2 imagery. The model unifies radar and optical inputs through modality-specific patch embeddings and adaptive cross-attention fusion. TerraFM achieves strong generalization on classification and segmentation tasks, outperforming prior models on GEO-Bench and Copernicus-Bench.
Researchers from MBZUAI have introduced VideoMolmo, a large multimodal model for spatio-temporal pointing conditioned on textual descriptions. The model incorporates a temporal module with an attention mechanism and a temporal mask fusion pipeline using SAM2 for improved coherence across video sequences. They also curated a dataset of 72k video-caption pairs and introduced VPoS-Bench, a benchmark for evaluating generalization across real-world scenarios, with code and models publicly available.
MBZUAI researchers introduce VideoMathQA, a new benchmark for evaluating mathematical reasoning in videos, requiring models to interpret visual information, text, and spoken cues. The dataset spans 10 mathematical domains with videos ranging from 10 seconds to over 1 hour, and includes multi-step reasoning annotations. The benchmark aims to evaluate temporal cross-modal reasoning and highlights the limitations of existing approaches in complex video-based mathematical problem solving.
This paper introduces a novel evaluation framework for Arabic language models, addressing gaps in linguistic accuracy and cultural alignment. The authors analyze existing datasets and present the Arabic Depth Mini Dataset (ADMD), a curated collection of 490 questions across ten domains. Evaluating GPT-4, Claude 3.5 Sonnet, Gemini Flash 1.5, CommandR 100B, and Qwen-Max using ADMD reveals performance variations, with Claude 3.5 Sonnet achieving the highest accuracy at 30%. Why it matters: The work emphasizes the importance of cultural competence in Arabic language model evaluation, providing practical insights for improvement.
MBZUAI introduces Agent-X, a benchmark for evaluating multi-step reasoning in vision-centric agents across real-world, multimodal settings. Agent-X includes 828 tasks with diverse visual contexts and spans six environments, requiring tool use and stepwise decision-making. Experiments show that current LLMs struggle with multi-step vision tasks, achieving less than 50% success, highlighting areas for improvement in LMM reasoning and tool use.
This paper analyzes Arabic text generated by LLMs like ALLaM, Jais, Llama, and GPT-4 across academic and social media domains using stylometric analysis. The study found detectable linguistic patterns that differentiate human-written from machine-generated Arabic text. BERT-based detection models achieved up to 99.9% F1-score in formal contexts, though cross-domain generalization remains a challenge. Why it matters: The research lays groundwork for detecting AI-generated misinformation in Arabic, a crucial step for preserving information integrity in Arabic-language contexts.
MBZUAI researchers introduce SocialMaze, a new benchmark for evaluating social reasoning capabilities in large language models (LLMs). SocialMaze includes six diverse tasks across social reasoning games, daily-life interactions, and digital community platforms, emphasizing deep reasoning, dynamic interaction, and information uncertainty. Experiments show that LLMs vary in handling dynamic interactions, degrade under uncertainty, but can be improved via fine-tuning on curated reasoning examples.
The Qatar Computing Research Institute (QCRI) has released SpokenNativQA, a multilingual spoken question-answering dataset for evaluating LLMs in conversational settings. The dataset contains 33,000 naturally spoken questions and answers across multiple languages, including low-resource and dialect-rich languages. It aims to address the limitations of text-based QA datasets by incorporating speech variability, accents, and linguistic diversity. Why it matters: This benchmark enables more robust evaluation of LLMs in speech-based interactions, particularly for Arabic dialects and other low-resource languages.
MBZUAI researchers release 'Fann or Flop', a new benchmark for evaluating Arabic poetry understanding in LLMs. The benchmark covers 12 historical eras and 14 poetic genres, assessing semantic understanding, metaphor interpretation, and cultural context. Evaluation of state-of-the-art LLMs reveals challenges in poetic understanding despite strong performance on standard Arabic benchmarks.
MBZUAI researchers introduce ARB, the first comprehensive benchmark for evaluating step-by-step multimodal reasoning in Arabic across textual and visual modalities. The benchmark spans 11 diverse domains and includes 1,356 multimodal samples with 5,119 human-curated reasoning steps. Evaluations of 12 state-of-the-art LMMs revealed challenges in coherence, faithfulness, and cultural grounding, highlighting the need for culturally aware AI systems.
Researchers from MBZUAI have introduced UrduFactCheck, a new framework for fact-checking in Urdu, along with two datasets: UrduFactBench and UrduFactQA. The framework uses monolingual and translation-based evidence retrieval to address the lack of Urdu resources. Evaluations using twelve LLMs showed that translation-augmented methods improve performance, highlighting challenges for open-source LLMs in Urdu.
MBZUAI researchers introduce FAID, a fine-grained AI-generated text detection framework capable of classifying text as human-written, LLM-generated, or collaboratively written. FAID utilizes multi-level contrastive learning and multi-task auxiliary classification to capture authorship and model-specific characteristics, and can identify the underlying LLM family. The framework outperforms existing baselines, especially in generalizing to unseen domains and new LLMs, and includes a multilingual, multi-domain dataset called FAIDSet.
MBZUAI researchers introduce LLM-BabyBench, a benchmark suite for evaluating grounded planning and reasoning in LLMs. The suite, built on a textual adaptation of the BabyAI grid world, assesses LLMs on predicting action consequences, generating action sequences, and decomposing instructions. Datasets, evaluation harness, and metrics are publicly available to facilitate reproducible assessment.
The United Arab Emirates has reportedly made a significant advancement in its efforts to compete in the global large language model (LLM) development landscape. This initiative signifies a strategic move by the UAE to establish itself as a key player in frontier AI research and innovation. This 'bold leap' likely encompasses new investments, research programs, or the release of an advanced LLM. Why it matters: This development underscores the UAE's ambition to leverage AI for economic diversification and technological leadership in the Middle East and globally.
This paper introduces SemDiff, a novel method for generating unrestricted adversarial examples (UAEs) by exploring the semantic latent space of diffusion models. SemDiff uses multi-attribute optimization to ensure attack success while preserving the naturalness and imperceptibility of generated UAEs. Experiments on high-resolution datasets demonstrate SemDiff's superior performance compared to state-of-the-art methods in attack success rate and imperceptibility, while also evading defenses.
The paper introduces SaudiCulture, a new benchmark for evaluating the cultural competence of LLMs within Saudi Arabia, covering five major geographical regions and diverse cultural domains. The benchmark includes questions of varying complexity and distinguishes between common and specialized regional knowledge. Evaluations of five LLMs (GPT-4, Llama 3.3, FANAR, Jais, and AceGPT) revealed performance declines on region-specific questions, highlighting the need for region-specific knowledge in LLM training.
Researchers introduce SALT, a parameter-efficient fine-tuning method for medical image segmentation that combines singular value adaptation with low-rank transformation. SALT selectively adapts influential singular values and complements this with a low-rank update for the remaining subspace. Experiments on five medical datasets show SALT outperforms state-of-the-art PEFT methods by 2-5% in Dice score with only 3.9% trainable parameters.
The paper introduces UAE-3D, a multi-modal VAE for 3D molecule generation that compresses molecules into a unified latent space, maintaining near-zero reconstruction error. This approach simplifies latent diffusion modeling by eliminating the need to handle multi-modality and equivariance separately. Experiments on GEOM-Drugs and QM9 datasets show UAE-3D establishes new benchmarks in de novo and conditional 3D molecule generation, with significant improvements in efficiency and quality.
Researchers at MBZUAI have introduced a novel approach to enhance Large Multimodal Models (LMMs) for autonomous driving by integrating 3D tracking information. This method uses a track encoder to embed spatial and temporal data, enriching visual queries and improving the LMM's understanding of driving scenarios. Experiments on DriveLM-nuScenes and DriveLM-CARLA benchmarks demonstrate significant improvements in perception, planning, and prediction tasks compared to baseline models.
MBZUAI researchers introduce LLMVoX, a 30M-parameter, LLM-agnostic, autoregressive streaming text-to-speech (TTS) system that generates high-quality speech with low latency. The system preserves the capabilities of the base LLM and achieves a lower Word Error Rate compared to speech-enabled LLMs. LLMVoX supports seamless, infinite-length dialogues and generalizes to new languages with dataset adaptation, including Arabic.
A new survey paper provides a deep dive into post-training methodologies for Large Language Models (LLMs), analyzing their role in refining LLMs beyond pretraining. It addresses key challenges such as catastrophic forgetting, reward hacking, and inference-time trade-offs, and highlights emerging directions in model alignment, scalable adaptation, and inference-time reasoning. The paper also provides a public repository to continually track developments in this fast-evolving field.
A new culturally inclusive and linguistically diverse dataset called Palm for Arabic LLMs is introduced, covering 22 Arab countries and featuring instructions in both Modern Standard Arabic (MSA) and dialectal Arabic (DA) across 20 topics. The dataset was built through a year-long community-driven project involving 44 researchers from across the Arab world. Evaluation of frontier LLMs using the dataset reveals limitations in cultural and dialectal understanding, with some countries being better represented than others.
Researchers introduce a benchmark to evaluate the factual recall and knowledge transferability of multilingual language models across 13 languages. The study reveals that language models often fail to transfer knowledge between languages, even when they possess the correct information in one language. The benchmark and evaluation framework are released to drive future research in multilingual knowledge transfer.
Researchers introduce TimeTravel, a benchmark dataset for evaluating large multimodal models (LMMs) on historical and cultural artifacts. The benchmark comprises 10,250 expert-verified samples across 266 cultures and 10 historical regions, designed to assess AI in tasks like classification and interpretation of manuscripts, artworks, inscriptions, and archaeological discoveries. The goal is to establish AI as a reliable partner in preserving cultural heritage and assisting researchers.
A new dataset called ArabCulture is introduced to address the lack of culturally relevant commonsense reasoning resources in Arabic AI. The dataset covers 13 countries across the Gulf, Levant, North Africa, and the Nile Valley, spanning 12 daily life domains with 54 fine-grained subtopics. It was built from scratch by native speakers writing and validating culturally relevant questions. Why it matters: The dataset highlights the need for more culturally aware models and benchmarks tailored to the Arabic-speaking world, moving beyond machine-translated resources.
The paper introduces MultiProSE, the first multi-label Arabic dataset for propaganda, sentiment, and emotion detection. It extends the existing ArPro dataset with sentiment and emotion annotations, resulting in 8,000 annotated news articles. Baseline models, including GPT-4o-mini and BERT-based models, were developed for each task, and the dataset, guidelines, and code are publicly available. Why it matters: This resource enables further research into Arabic language models and a better understanding of opinion dynamics within Arabic news media.
This research evaluates LLMs like ChatGPT, Llama, Aya, Jais, and ACEGPT on Arabic automated essay scoring (AES) using the AR-AES dataset. The study uses zero-shot, few-shot learning, and fine-tuning approaches while using a mixed-language prompting strategy. ACEGPT performed best among the LLMs with a QWK of 0.67, while a smaller BERT model achieved 0.88. Why it matters: The study highlights challenges faced by LLMs in processing Arabic and provides insights into improving LLM performance in Arabic NLP tasks.
Hamad Bin Khalifa University's Qatar Computing Research Institute (QCRI) introduced Fanar, an Arabic-centric multimodal generative AI platform featuring the Fanar Star (7B) and Fanar Prime (9B) Arabic LLMs. These models were trained on nearly 1 trillion tokens and are designed to address different prompts through a custom orchestrator. Fanar includes a customized Islamic RAG system, a Recency RAG, bilingual speech recognition, and an attribution service for content verification, sponsored by Qatar's Ministry of Communications and Information Technology. Why it matters: The platform signifies a major step towards sovereign AI development in Qatar, providing advanced Arabic language capabilities and addressing regional needs.
The paper introduces VENOM, a text-driven framework for generating high-quality unrestricted adversarial examples using diffusion models. VENOM unifies image content generation and adversarial synthesis into a single reverse diffusion process, enhancing both attack success rate and image quality. The framework incorporates an adaptive adversarial guidance strategy with momentum to ensure the generated adversarial examples align with the distribution of natural images.
This paper introduces AraLLaMA, a new Arabic large language model (LLM) trained using a progressive vocabulary expansion method inspired by second language acquisition. The model utilizes a modified byte-pair encoding (BPE) algorithm to dynamically extend the Arabic subwords in its vocabulary during training, balancing the out-of-vocabulary (OOV) ratio. Experiments show AraLLaMA achieves performance comparable to existing Arabic LLMs on various benchmarks, and all models, data, and code will be open-sourced. Why it matters: This work addresses the need for more accessible and performant Arabic LLMs, contributing to democratization of AI in the Arab world.
Researchers at MBZUAI have demonstrated a method called "Data Laundering" to artificially boost language model benchmark scores using knowledge distillation. The technique covertly transfers benchmark-specific knowledge, leading to inflated accuracy without genuine improvements in reasoning. The study highlights a vulnerability in current AI evaluation practices and calls for more robust benchmarks.
MBZUAI researchers introduce UniMed-CLIP, a unified Vision-Language Model (VLM) for diverse medical imaging modalities, trained on the new large-scale, open-source UniMed dataset. UniMed comprises over 5.3 million image-text pairs across six modalities: X-ray, CT, MRI, Ultrasound, Pathology, and Fundus, created using LLMs to transform classification datasets into image-text formats. UniMed-CLIP significantly outperforms existing generalist VLMs and matches modality-specific medical VLMs in zero-shot evaluations, improving over BiomedCLIP by +12.61 on average across 21 datasets while using 3x less training data.
MBZUAI releases BiMediX2, a bilingual (Arabic-English) Bio-Medical Large Multimodal Model, along with the BiMed-V dataset (1.6M samples) and BiMed-MBench evaluation benchmark. BiMediX2 supports multi-turn conversation in Arabic and English and handles diverse medical imaging modalities. The model achieves state-of-the-art results on medical LLM and LMM benchmarks, outperforming existing methods and GPT-4 in specific evaluations.
The paper introduces Arabic Stable LM, a 1.6B parameter Arabic-centric language model, in both base and chat versions. The Arabic Stable LM 1.6B chat model achieves strong results on several benchmarks, outperforming models with up to 8x more parameters. The study also demonstrates the benefit of incorporating synthetic instruction tuning data through a large synthetic dialogue dataset. Why it matters: This work makes Arabic LLMs more accessible by reducing the parameter size while maintaining strong performance, facilitating deployment in resource-constrained environments.
A new study introduces Sporo AraSum, a language model designed for Arabic clinical documentation, and compares it to JAIS using synthetic datasets and modified PDQI-9 metrics. Sporo AraSum significantly outperformed JAIS in quantitative AI metrics and qualitative attributes related to accuracy, utility, and cultural competence. The model addresses the nuances of Arabic while reducing AI hallucinations, making it suitable for Arabic-speaking healthcare. Why it matters: The model offers a more culturally and linguistically sensitive solution for Arabic clinical documentation, potentially improving healthcare workflows and patient outcomes in the region.
Researchers introduce Swan, a family of Arabic-centric embedding models including Swan-Small (based on ARBERTv2) and Swan-Large (based on ArMistral). They also propose ArabicMTEB, a benchmark suite for cross-lingual, multi-dialectal Arabic text embedding performance across 8 tasks and 94 datasets. Swan-Large achieves state-of-the-art results, outperforming Multilingual-E5-large in most Arabic tasks. Why it matters: The new models and benchmarks address a critical need for high-quality Arabic language models that are both dialectally and culturally aware, enabling more effective NLP applications in the region.
Researchers have introduced LlamaLens, a specialized multilingual LLM designed for analyzing news and social media content. The model addresses domain specificity and multilinguality, with a focus on news and social media in Arabic, English, and Hindi. LlamaLens was evaluated on 18 tasks represented by 52 datasets, outperforming the state-of-the-art on 23 testing sets. Why it matters: This work contributes a valuable resource for multilingual NLP research, particularly in the context of analyzing news and social media content across diverse languages.
A novel agent-based framework called FIRE is introduced for fact-checking long-form text. FIRE iteratively integrates evidence retrieval and claim verification, deciding whether to provide a final answer or generate a subsequent search query. Experiments show FIRE achieves comparable performance to existing methods while reducing LLM costs by 7.6x and search costs by 16.5x.
Researchers developed Atlas-Chat, a collection of LLMs for dialectal Arabic, focusing on Moroccan Arabic (Darija). They constructed an instruction dataset by consolidating existing Darija language resources and translating English instructions. Atlas-Chat models (2B, 9B, 27B) outperform state-of-the-art and Arabic-specialized LLMs like LLaMa, Jais, and AceGPT on Darija NLP tasks. Why it matters: This work addresses the gap in LLM support for low-resource Arabic dialects, providing a methodology for instruction-tuning and benchmarks for future research.
The paper introduces Juhaina, a 9.24B parameter Arabic-English bilingual LLM trained with an 8,192 token context window. It identifies limitations in the Open Arabic LLM Leaderboard (OALL) and proposes a new benchmark, CamelEval, for more comprehensive evaluation. Juhaina outperforms models like Llama and Gemma in generating helpful Arabic responses and understanding cultural nuances. Why it matters: This culturally-aligned LLM and associated benchmark could significantly advance Arabic NLP and democratize AI access for Arabic speakers.
Researchers introduce AraDiCE, a benchmark for Arabic Dialect and Cultural Evaluation, comprising seven synthetic datasets in various dialects and Modern Standard Arabic (MSA). The benchmark includes approximately 45,000 post-edited samples and evaluates LLMs on dialect comprehension, generation, and cultural awareness across the Gulf, Egypt, and Levant. Results show that Arabic-specific models like Jais and AceGPT outperform multilingual models on dialectal tasks, but challenges remain in dialect identification, generation, and translation. Why it matters: This benchmark and associated datasets will help improve LLMs' ability to understand and generate diverse Arabic dialects and cultural contexts, addressing a significant gap in current models.
KAUST has established a Center of Excellence (CoE) for Generative AI, chaired by Professor Bernard Ghanem and co-chaired by Professor Jürgen Schmidhuber. The center will focus on scientific research, commercial innovation, and talent development in GenAI, aligning with Saudi Arabia's Vision 2030 goals. The CoE aims to impact Saudi Arabia's four RDI priorities: Health and Wellness, Sustainable Environment, Energy and Industrial Leadership, and Economies of the Future. Why it matters: The KAUST center aims to position Saudi Arabia as a global leader in generative AI, addressing the need for specialized expertise and infrastructure while contributing to the Kingdom's economic diversification.