Jais 2 is a family of open-weight, Arabic-centric large language models designed for strong Arabic performance and broad cultural coverage. The family includes, to our knowledge, the largest open Arabic-centric LLM trained from scratch, with 70B parameters, alongside an efficient 8B model. A custom Arabic-centric vocabulary supports efficient training and inference, while an optimized architecture and training recipe improve compute efficiency. Jais 2 achieves the highest scores on the Open Arabic LLM Leaderboard v2 (OALL2) and AraGen v1 among the evaluated open models, while remaining competitive on English benchmarks. It also obtains strong results in Arabic poetry, dream interpretation, translation, and summarization. On the IslamicQA domain, Jais 2-70B achieves the second-highest overall average among the evaluated large open models, while attaining the highest scores on PalmX and the held-out in-house IslamicQA benchmark. We release both model sizes as open-weights on Hugging Face under a commercially permissive license. Jais 2-70B is also available as a chat application on the Web, iOS, and Android, reaching inference speeds of up to 2,000 tokens per second on Cerebras hardware.

Large reasoning models (LRMs) produce reasoning traces (RTs) that often contain sensitive information. These leaky thoughts are difficult to control and frequently violate explicit privacy directives. Because RTs can be exposed through prompt injection attacks, this becomes a direct privacy risk to the user. We approach this as a controllability problem: since privacy directives are themselves instructions, improving instruction-following (IF) within the RT provides a direct path to reducing privacy leaks. To this end, we introduce an SFT dataset that teaches models to follow general instructions throughout their reasoning process, and propose Staged Decoding, a simple decoding strategy that decouples RT and answer generation using separate LoRA adapters to maximize IF of each component. We evaluate our approach on six models from two families (1.7B-14B parameters), across two IF benchmarks and two privacy benchmarks. Our method yields substantial improvements, with gains of up to 20.9 points in IF and 51.9 percentage points on privacy benchmarks, though these can come at the cost of task utility due to the trade-off between reasoning performance and IF. Our results show that improving IF in LRMs can significantly enhance privacy, suggesting a promising direction for future privacy-aware LRMs.

Long-context understanding has emerged as a critical capability for large language models (LLMs). However, evaluating this ability remains challenging. We present SCALAR, a benchmark designed to assess citation-grounded long-context reasoning in academic writing. SCALAR leverages academic papers and their citation structure to automatically generate high-quality ground-truth labels without human annotation. It features controllable difficulty levels and a dynamic updating mechanism that mitigates data contamination. The benchmark includes two tasks: a multiple-choice QA format and a cloze-style citation prediction. We evaluate a range of state-of-the-art LLMs and find that the multiple-choice task effectively distinguishes model capabilities—while human experts achieve over 90% accuracy, most models struggle. The cloze-style task is even more challenging, with no model exceeding 40% accuracy. SCALAR provides a domain-grounded, continuously updating framework for tracking progress in citation-based long-context understanding. Code and data will be publicly released.

Large language models remain predominantly English-centric, which limits their utility for underrepresented languages. We help bridge this gap for Hindi with Llama-3-Nanda-10B-Chat (aka Nanda-10B) and Llama-3.1-Nanda-87B-Chat (aka Nanda-87B), forming the Nanda family of open-weight bilingual models (https://github.com/MBZUAI-IFM/Nanda-Family). Our approach integrates: (i) a tokenizer extending Llama’s vocabulary with 20% Hindi-specific tokens, thus halving Hindi tokenization fertility while preserving English efficiency, (ii) Hindi-first parameter-efficient continual pretraining using Llama Pro on a 65B-token corpus spanning Devanagari script, code-mixed, and Romanized Hindi, and (iii) bilingual instruction and safety alignment on a large culturally grounded dataset. The resulting Nanda models outperform open-weight LLMs of comparable size: Nanda-87B yields high generative quality, and Nanda-10B shows competitive general-purpose performance. Nanda-87B demonstrates state-of-the-art performance on summarization, translation, transliteration, and instruction following. Moreover, both models achieve state-of-the-art performance in safety and in cultural knowledge. Our results demonstrate that careful tokenizer design, data curation, and continual pretraining can yield capable and safe LLMs for resource-poor languages without compromising English performance.

Multi-step symbolic reasoning is essential for robust financial analysis; yet, current benchmarks largely overlook this capability. Existing datasets such as FinQA and ConvFinQA emphasize final numerical answers while neglecting the intermediate reasoning steps required for transparency and verification. To address this gap, we introduce FinChain, the first benchmark specifically designed for verifiable Chain-of-Thought evaluation in finance. FinChain spans 58 topics across 12 financial domains, each represented by parameterized symbolic templates with executable Python code that enable fully machine-verifiable reasoning and scalable, contamination-free data generation.To assess reasoning capacity, we propose ChainEval, a dynamic alignment measure that jointly evaluates both the final-answer correctness and the step-level reasoning consistency. Our evaluation of 26 leading LLMs reveals that even frontier LLMs exhibit clear limitations in symbolic financial reasoning, while domain-adapted and math-enhanced fine-tuned models can substantially narrow this gap.Overall, FinChain exposes persistent weaknesses in multi-step financial reasoning and provides a foundation for developing trustworthy, interpretable, and verifiable financial AI. This project is available at https://github.com/mbzuai-nlp/finchain.git.

Idiomatic expressions pose a major challenge for multilingual NLP because their meanings shift between figurative and literal usage, often requiring context for accurate interpretation. Prior work has focused on high-resource languages typically evaluates isolated idiom-meaning questions, overlooking realistic discourse. We introduce MIDI, a multilingual idiom dataset spanning 3 high-, 3 medium-, and 12 low-resource languages, curated by native speakers. Unlike previous datasets, MIDI provides idioms embedded in both sentence-level and conversational contexts, capturing both literal and figurative readings. Benchmarking state-of-the-art models shows that idiom comprehension degrades in low-resource languages and that, in all resource tiers, literal interpretations are substantially harder than figurative ones. Conversational context improves performance but does not eliminate these disparities. Through controlled tests and interventions on hidden representations, we further separate memorization from reasoning, exposing core limitations of current models.

Code retrieval is vital to modern software engineering as it boosts reuse and speeds up debugging. However, current benchmarks primarily emphasize functional relevance while neglecting code quality. To address this gap, we introduce CoQuIR, the first large-scale, multilingual benchmark specifically designed to evaluate quality-aware code retrieval across four critical dimensions: correctness, efficiency, security, and maintainability. CoQuIR includes fine-grained quality annotations over 42,725 queries and 134,907 code snippets in 11 programming languages. Evaluating 23 retrievers (both open-source and proprietary) shows that even state-of-the-art models often fail to separate buggy or insecure code from robust counterparts. We further investigate methods for explicitly training retrievers to recognize code quality, demonstrating that quality-aware metrics can be improved without loss of semantic relevance; downstream code generation benefits from these gains. CoQuIR underscores the importance of embedding quality signals into retrieval systems as a crucial component for more trustworthy developer tools.

LLMs are expected to respond consistently across demographic groups, yet this assumption remains largely untested due to the absence of demographic information in existing instruction datasets. To address this gap, we introduce PromptDial, a collection of 2,289 English prompts written by real users and annotated with seven demographic attributes: sex, race, education, age, language, employment sector, and nationality. We evaluate state-of-the-art generative models on 39 datasets, including machine translation, summarization, grammar correction, knowledge and reasoning, semantics, and question answering, and observe performance disparities of up to 7.7% between demographic groups, with statistically significant differences in over half of the datasets. Our linguistic analysis points to variation in prompt tone and linguistic features as potential drivers of these disparities. Our findings suggest that current instruction tuning practices overlook key aspects of linguistic diversity, and we call for the inclusion of demographic metadata and more representative prompt data to support fairer and more robust language model behavior.

Interaction trajectories between agents and environments have proven effective in tuning LLMs into task-specific agents. However, constructing these trajectories, especially successful trajectories, is often computationally and time intensive due to the relatively low success rates of even the most advanced LLMs, such as GPT-4 and Claude. Additionally, common training paradigms like supervised fine-tuning (SFT) and reinforcement learning (RL) not only require large volumes of data but also have specific demands regarding the trajectories used. For instance, existing SFT approaches typically utilize only positive examples, limiting their efficiency in low-resource scenarios. To address this, we introduce Negative-Aware Training (NAT), a straightforward yet effective method that leverages both successful and failed trajectories for fine-tuning, maximizing the utility of limited resources. Experimental results demonstrate that NAT consistently surpasses existing methods, including SFT, DPO, and PPO, across various tasks.

The impressive advancement of Large Language Models (LLMs) in English has not been matched across all languages. In particular, LLM performance in Arabic lags behind, due to data scarcity, linguistic diversity of Arabic and its dialects, morphological complexity, etc. Progress is further hindered by the quality of Arabic benchmarks, which typically rely on static, publicly available data, lack comprehensive task coverage, or do not provide dedicated platforms with blind test sets. This makes it challenging to measure actual progress and to mitigate data contamination. Here, we aim to bridge these gaps. In particular, we introduce BALSAM, a comprehensive, community-driven benchmark aimed at advancing Arabic LLM development and evaluation. It includes 78 NLP tasks from 14 broad categories, with 52K examples divided into 37K test and 15K development, and a centralized, transparent platform for blind evaluation. We envision BALSAM as a unifying platform that sets standards and promotes collaborative research to advance Arabic LLM capabilities.

With the rapid evolution of large language models (LLMs), new and hard-to-predict harmful capabilities are emerging. This requires developers to identify potential risks through the evaluation of “dangerous capabilities” in order to responsibly deploy LLMs. Here we aim to facilitate this process. In particular, we collect an open-source dataset to evaluate the safeguards in LLMs, to facilitate the deployment of safer open-source LLMs at a low cost. Our dataset is curated and filtered to consist only of instructions that responsible language models should not follow. We assess the responses of six popular LLMs to these instructions, and we find that simple BERT-style classifiers can achieve results that are comparable to GPT-4 on automatic safety evaluation. Our data and code are available at https://github.com/Libr-AI/do-not-answer

As the capabilities of large language models (LLMs) continue to advance, evaluating their performance is becoming more important and more challenging. This paper aims to address this issue for Mandarin Chinese in the form of CMMLU, a comprehensive Chinese benchmark that covers various subjects, including natural sciences, social sciences, engineering, and the humanities. We conduct a thorough evaluation of more than 20 contemporary multilingual and Chinese LLMs, assessing their performance across different subjects and settings. The results reveal that most existing LLMs struggle to achieve an accuracy of even 60%, which is the pass mark for Chinese exams. This highlights that there is substantial room for improvement in the capabilities of LLMs. Additionally, we conduct extensive experiments to identify factors impacting the models’ performance and propose directions for enhancing LLMs. CMMLU fills the gap in evaluating the knowledge and reasoning capabilities of large language models for Chinese.

Large language models (LLMs) are notorious for hallucinating, i.e., producing erroneous claims in their output. Such hallucinations can be dangerous, as occasional factual inaccuracies in the generated text might be obscured by the rest of the output being generally factually correct, making it extremely hard for the users to spot them. Current services that leverage LLMs usually do not provide any means for detecting unreliable generations. Here, we aim to bridge this gap. In particular, we propose a novel fact-checking and hallucination detection pipeline based on token-level uncertainty quantification. Uncertainty scores leverage information encapsulated in the output of a neural network or its layers to detect unreliable predictions, and we show that they can be used to fact-check the atomic claims in the LLM output. Moreover, we present a novel token-level uncertainty quantification method that removes the impact of uncertainty about what claim to generate on the current step and what surface form to use. Our method Claim Conditioned Probability (CCP) measures only the uncertainty of a particular claim value expressed by the model. Experiments on the task of biography generation demonstrate strong improvements for CCP compared to the baselines for seven different LLMs and four languages. Human evaluation reveals that the fact-checking pipeline based on uncertainty quantification is competitive with a fact-checking tool that leverages external knowledge.

The focus of language model evaluation has transitioned towards reasoning and knowledge-intensive tasks, driven by advancements in pretraining large models. While state-of-the-art models are partially trained on large Arabic texts, evaluating their performance in Arabic remains challenging due to the limited availability of relevant datasets. To bridge this gap, we present ArabicMMLU, the first multi-task language understanding benchmark for the Arabic language, sourced from school exams across diverse educational levels in different countries spanning North Africa, the Levant, and the Gulf regions. Our data comprises 40 tasks and 14,575 multiple-choice questions in Modern Standard Arabic (MSA) and is carefully constructed by collaborating with native speakers in the region. Our comprehensive evaluations of 35 models reveal substantial room for improvement, particularly among the best open-source models. Notably, BLOOMZ, mT0, LLama2, and Falcon struggle to achieve a score of 50%, while even the top-performing Arabic-centric model only achieves a score of 62.3%.

Many studies have demonstrated that large language models (LLMs) can produce harmful responses, exposing users to unexpected risks. Previous studies have proposed comprehensive taxonomies of LLM risks, as well as corresponding prompts that can be used to examine LLM safety. However, the focus has been almost exclusively on English. We aim to broaden LLM safety research by introducing a dataset for the safety evaluation of Chinese LLMs, and extending it to better identify false negative and false positive examples in terms of risky prompt rejections. We further present a set of fine-grained safety assessment criteria for each risk type, facilitating both manual annotation and automatic evaluation in terms of LLM response harmfulness. Our experiments over five LLMs show that region-specific risks are the prevalent risk type. Warning: this paper contains example data that may be offensive, harmful, or biased. Our data is available at https://github.com/Libr-AI/do-not-answer.

Instruction tuning significantly enhances the performance of large language models (LLMs) across various tasks. However, the procedure to optimizing the mixing of instruction datasets for LLM fine-tuning is still poorly understood. This study categorizes instructions into three primary types: NLP downstream tasks, coding, and general chat. We explore the effects of instruction tuning on different combinations of datasets on LLM performance, and find that certain instruction types are more advantageous for specific applications but can negatively impact other areas. This work provides insights into instruction mixtures, laying the foundations for future research.

Most existing reading comprehension datasets focus on single-span answers, which can be extracted as a single contiguous span from a given text passage. Multi-span questions, i.e., questions whose answer is a series of multiple discontiguous spans in the text, are common real life but are less studied. In this paper, we present MultiSpanQA, a new dataset that focuses on multi-span questions. Raw questions and contexts are extracted from the Natural Questions dataset. After multi-span re-annotation, MultiSpanQA consists of over a total of 6,000 multi-span questions in the basic version, and over 19,000 examples with unanswerable questions, and questions with single-, and multi-span answers in the expanded version. We introduce new metrics for the purposes of multi-span question answering evaluation, and establish several baselines using advanced models. Finally, we propose a new model which beats all baselines and achieves state-of-the-art on our dataset.

Pre-trained language models (PLMs) have dramatically improved performance for many natural language processing (NLP) tasks in domains such as finance and healthcare. However, the application of PLMs in the domain of commerce, especially marketing and advertising, remains less studied. In this work, we adapt pre-training methods to the domain of commerce, by proposing CULG, a large-scale commercial universal language generation model which is pre-trained on a corpus drawn from 10 markets across 7 languages. We propose 4 commercial generation tasks and a two-stage training strategy for pre-training, and demonstrate that the proposed strategy yields performance improvements on three generation tasks as compared to single-stage pre-training. Extensive experiments show that our model outperforms other models by a large margin on commercial generation tasks, and we conclude with a discussion on additional applications over other markets, languages, and tasks.