Akshay Nambi

Kabir Ahuja, Harshita Diddee, Rishav Hada et al. · microsoft-research

Generative AI models have shown impressive performance on many Natural Language Processing tasks such as language understanding, reasoning, and language generation. An important question being asked by the AI community today is about the capabilities and limits of these models, and it is clear that evaluating generative AI is very challenging. Most studies on generative LLMs have been restricted to English and it is unclear how capable these models are at understanding and generating text in other languages. We present the first comprehensive benchmarking of generative LLMs - MEGA, which evaluates models on standard NLP benchmarks, covering 16 NLP datasets across 70 typologically diverse languages. We compare the performance of generative LLMs including Chat-GPT and GPT-4 to State of the Art (SOTA) non-autoregressive models on these tasks to determine how well generative models perform compared to the previous generation of LLMs. We present a thorough analysis of the performance of models across languages and tasks and discuss challenges in improving the performance of generative LLMs on low-resource languages. We create a framework for evaluating generative LLMs in the multilingual setting and provide directions for future progress in the field.

Nirmit Arora, Sathvik Joel, Ishan Kavathekar et al.

LLM-based agents are increasingly deployed in multi-agent systems (MAS). As these systems move toward real-world applications, their security becomes paramount. Existing research largely evaluates single-agent security, leaving a critical gap in understanding the vulnerabilities introduced by multi-agent design. However, existing systems fall short due to lack of unified frameworks and metrics focusing on unique rejection modes in MAS. We present SafeAgents, a unified and extensible framework for fine-grained security assessment of MAS. SafeAgents systematically exposes how design choices such as plan construction strategies, inter-agent context sharing, and fallback behaviors affect susceptibility to adversarial prompting. We introduce Dharma, a diagnostic measure that helps identify weak links within multi-agent pipelines. Using SafeAgents, we conduct a comprehensive study across five widely adopted multi-agent architectures (centralized, decentralized, and hybrid variants) on four datasets spanning web tasks, tool use, and code generation. Our findings reveal that common design patterns carry significant vulnerabilities. For example, centralized systems that delegate only atomic instructions to sub-agents obscure harmful objectives, reducing robustness. Our results highlight the need for security-aware design in MAS. Link to code is https://github.com/microsoft/SafeAgents

Namita Singh, Jacqueline Wang'ombe, Nereah Okanga et al.

Small and medium-sized agricultural holders face challenges like limited access to localized, timely information, impacting productivity and sustainability. Traditional extension services, which rely on in-person agents, struggle with scalability and timely delivery, especially in remote areas. We introduce FarmerChat, a generative AI-powered chatbot designed to address these issues. Leveraging Generative AI, FarmerChat offers personalized, reliable, and contextually relevant advice, overcoming limitations of previous chatbots in deterministic dialogue flows, language support, and unstructured data processing. Deployed in four countries, FarmerChat has engaged over 15,000 farmers and answered over 300,000 queries. This paper highlights how FarmerChat's innovative use of GenAI enhances agricultural service scalability and effectiveness. Our evaluation, combining quantitative analysis and qualitative insights, highlights FarmerChat's effectiveness in improving farming practices, enhancing trust, response quality, and user engagement.

Aradhye Agarwal, Gurdit Siyan, Yash Pandya et al.

Agentic language models operate in a fundamentally different safety regime than chat models: they must plan, call tools, and execute long-horizon actions where a single misstep, such as accessing files or entering credentials, can cause irreversible harm. Existing alignment methods, largely optimized for static generation and task completion, break down in these settings due to sequential decision-making, adversarial tool feedback, and overconfident intermediate reasoning. We introduce MOSAIC, a post-training framework that aligns agents for safe multi-step tool use by making safety decisions explicit and learnable. MOSAIC structures inference as a plan, check, then act or refuse loop, with explicit safety reasoning and refusal as first-class actions. To train without trajectory-level labels, we use preference-based reinforcement learning with pairwise trajectory comparisons, which captures safety distinctions often missed by scalar rewards. We evaluate MOSAIC zero-shot across three model families, Qwen2.5-7B, Qwen3-4B-Thinking, and Phi-4, and across out-of-distribution benchmarks spanning harmful tasks, prompt injection, benign tool use, and cross-domain privacy leakage. MOSAIC reduces harmful behavior by up to 50%, increases harmful-task refusal by over 20% on injection attacks, cuts privacy leakage, and preserves or improves benign task performance, demonstrating robust generalization across models, domains, and agentic settings.

Ashutosh Bajpai, Tamal Majumder, Akshay Nambi et al.

Small Vision-Language Models (SVLMs) are efficient task controllers but often suffer from visual brittleness and poor tool orchestration. They typically require expensive supervised trajectory tuning to mitigate these deficits. In this work, we propose Self-supervised Perception Enabled by Cascaded Tool Rollout Alignment (SPECTRA), a supervision-free framework that bootstraps agentic capabilities via Coldstart Reinforcement Learning for SVLMs. SPECTRA enforces Soft Structured Multi-turn Rollouts, a topological constraint that directs agents to explicitly sequence tool derived evidence before synthesis, effectively grounding reasoning in visual observations. We employ a multi-objective reward signal that simultaneously maximizes task correctness, rollout structure, and tool utility, enabling agent to self-discover robust behaviors without human preference labels. We further introduce Tool Instrumental Utility (TIU), a novel metric to quantify tool efficacy in the absence of ground truth. Extensive evaluations across composite and out-of-distribution (MMMU-Pro) benchmarks demonstrate that SPECTRA boosts agentic trajectories, improving task accuracy by up to 5% and tool efficiency by 9%, enabling more efficient multimodal agents that learn effectively from environmental interaction alone.

Joykirat Singh, Zaid Khan, Archiki Prasad et al.

Large language models (LLMs) are increasingly deployed on long-horizon tasks in partially observable environments, where they must act while inferring and tracking a complex environment state over many steps. This leads to two challenges: partial observability requires maintaining uncertainty over unobserved world attributes, and long interaction history causes context to grow without bound, diluting task-relevant information. A principled solution to both challenges is a belief state: a posterior distribution over environment states given past observations and actions, which compactly encodes history for decision making regardless of episode length. In LLM agents, however, the open-ended nature of text makes it unclear how to represent such a distribution. Therefore, we introduce Agent-BRACE: Agent Belief state Representation via Abstraction and Confidence Estimation, a method that decouples an LLM agent into a belief state model and a policy model, jointly optimized via reinforcement learning. The belief state model produces a structured approximation of the belief distribution: a set of atomic natural language claims about the environment, each annotated with an ordinal verbalized certainty label ranging from certain to unknown. The policy model conditions on this compact, structured approximate belief rather than the full history, learning to select actions under explicit uncertainty. Across long-horizon, partially observable embodied language environments, Agent-BRACE achieves an average absolute improvement of +14.5% (Qwen2.5-3B-Instruct) and +5.3% (Qwen3-4B-Instruct), outperforming strong RL baselines while maintaining a near-constant context window independent of episode length. Further analysis shows that the learned belief becomes increasingly calibrated over the course of an episode as evidence accumulates.

Sidharth Sinha, Anson Bastos, Xuchao Zhang et al.

Large language models (LLMs) excel in open domains but struggle in specialized settings with limited data and evolving knowledge. Existing domain adaptation practices rely heavily on manual trial-and-error processes, incur significant hyperparameter complexity, and are highly sensitive to data and user preferences, all under the high cost of LLM training. Moreover, the interactions and transferability of hyperparameter choices across models/domains remain poorly understood, making adaptation gains uncertain even with substantial effort. To solve these challenges, we present AutoAdapt, a novel end-to-end automated framework for efficient and reliable LLM domain adaptation. AutoAdapt leverages curated knowledge bases from literature and open-source resources to reduce expert intervention. To narrow the search space, we design a novel multi-agent debating system in which proposal and critic agents iteratively interact to align user intent and incorporate data signals and best practices into the planning process. To optimize hyperparameters under tight budgets, we propose AutoRefine, a novel LLM-based surrogate that replaces costly black-box search. Across 10 tasks, AutoAdapt achieves a 25% average relative accuracy improvement over state-of-the-art Automated Machine Learning baselines with minimal overhead.

Nemin Wu, Qian Cao, Zhangyu Wang et al.

Spatial representation learning (SRL) aims at learning general-purpose neural network representations from various types of spatial data (e.g., points, polylines, polygons, networks, images, etc.) in their native formats. Learning good spatial representations is a fundamental problem for various downstream applications such as species distribution modeling, weather forecasting, trajectory generation, geographic question answering, etc. Even though SRL has become the foundation of almost all geospatial artificial intelligence (GeoAI) research, we have not yet seen significant efforts to develop an extensive deep learning framework and benchmark to support SRL model development and evaluation. To fill this gap, we propose TorchSpatial, a learning framework and benchmark for location (point) encoding, which is one of the most fundamental data types of spatial representation learning. TorchSpatial contains three key components: 1) a unified location encoding framework that consolidates 15 commonly recognized location encoders, ensuring scalability and reproducibility of the implementations; 2) the LocBench benchmark tasks encompassing 7 geo-aware image classification and 10 geo-aware image regression datasets; 3) a comprehensive suite of evaluation metrics to quantify geo-aware model's overall performance as well as their geographic bias, with a novel Geo-Bias Score metric. Finally, we provide a detailed analysis and insights into the model performance and geographic bias of different location encoders. We believe TorchSpatial will foster future advancement of spatial representation learning and spatial fairness in GeoAI research. The TorchSpatial model framework and LocBench benchmark are available at https://github.com/seai-lab/TorchSpatial, and the Geo-Bias Score evaluation framework is available at https://github.com/seai-lab/PyGBS.

Hengyi Wang, Haizhou Shi, Shiwei Tan et al.

Multimodal Large Language Models (MLLMs) have shown significant promise in various applications, leading to broad interest from researchers and practitioners alike. However, a comprehensive evaluation of their long-context capabilities remains underexplored. To address these gaps, we introduce the MultiModal Needle-in-a-haystack (MMNeedle) benchmark, specifically designed to assess the long-context capabilities of MLLMs. Besides multi-image input, we employ image stitching to further increase the input context length, and develop a protocol to automatically generate labels for sub-image level retrieval. Essentially, MMNeedle evaluates MLLMs by stress-testing their capability to locate a target sub-image (needle) within a set of images (haystack) based on textual instructions and descriptions of image contents. This setup necessitates an advanced understanding of extensive visual contexts and effective information retrieval within long-context image inputs. With this benchmark, we evaluate state-of-the-art MLLMs, encompassing both API-based and open-source models. The findings reveal that GPT-4o consistently surpasses other models in long-context scenarios, but suffers from hallucination problems in negative samples, i.e., when needles are not in the haystacks. Our comprehensive long-context evaluation of MLLMs also sheds lights on the considerable performance gap between API-based and open-source models. All the code, data, and instructions required to reproduce the main results are available at https://github.com/Wang-ML-Lab/multimodal-needle-in-a-haystack.

Joykirat Singh, Raghav Magazine, Yash Pandya et al.

Large language models (LLMs) have achieved remarkable progress in complex reasoning tasks, yet they remain fundamentally limited by their reliance on static internal knowledge and text-only reasoning. Real-world problem solving often demands dynamic, multi-step reasoning, adaptive decision making, and the ability to interact with external tools and environments. In this work, we introduce ARTIST (Agentic Reasoning and Tool Integration in Self-improving Transformers), a unified framework that tightly couples agentic reasoning, reinforcement learning, and tool integration for LLMs. ARTIST enables models to autonomously decide when, how, and which tools to invoke within multi-turn reasoning chains, leveraging outcome-based RL to learn robust strategies for tool use and environment interaction without requiring step-level supervision. Extensive experiments on mathematical reasoning and multi-turn function calling benchmarks show that ARTIST consistently outperforms state-of-the-art baselines, with up to 22% absolute improvement over base models and strong gains on the most challenging tasks. Detailed studies and metric analyses reveal that agentic RL training leads to deeper reasoning, more effective tool use, and higher-quality solutions. Our results establish agentic RL with tool integration as a powerful new frontier for robust, interpretable, and generalizable problem-solving in LLMs.

Joykirat Singh, Tanmoy Chakraborty, Akshay Nambi

Large language models (LLMs) have significantly improved their reasoning capabilities; however, they still struggle with complex multi-step mathematical problem-solving due to error propagation, lack of self-correction, and limited adaptability to diverse reasoning styles. Existing methods rely on static fine-tuning or prompt engineering, which fail to generalize across problem complexities, while the scarcity of high-quality preference data further hinders reliable reasoning. We introduce SPHERE, a self-evolving data generation pipeline that enhances reasoning in small language models (SLMs) by iteratively generating, correcting, and diversifying reasoning chains. SPHERE operates in three stages: (i) Self-Generation, where the model autonomously constructs problem-solving steps; (ii) Self-Correction, enabling it to identify and rectify errors; and (iii) Diversity Induction, improving robustness through multiple valid reasoning trajectories. This self-evolution mechanism strengthens mathematical reasoning and enhances model reliability. Evaluations on MATH 500, GSM8K, AIME, AMC, and Olympiad show that SPHERE-trained models achieve significant gains over their base versions and match/surpass GPT-4o on certain benchmarks. Our findings demonstrate that self-evolving models can close the reasoning gap between SLMs and state-of-the-art LLMs, making mathematical AI more reliable, scalable, and efficient.

Zhangyu Wang, Nemin Wu, Qian Cao et al.

The widespread adoption of AI models, especially foundation models (FMs), has made a profound impact on numerous domains. However, it also raises significant ethical concerns, including bias issues. Although numerous efforts have been made to quantify and mitigate social bias in AI models, geographic bias (in short, geo-bias) receives much less attention, which presents unique challenges. While previous work has explored ways to quantify geo-bias, these measures are model-specific (e.g., mean absolute deviation of LLM ratings) or spatially implicit (e.g., average fairness scores of all spatial partitions). We lack a model-agnostic, universally applicable, and spatially explicit geo-bias evaluation framework that allows researchers to fairly compare the geo-bias of different AI models and to understand what spatial factors contribute to the geo-bias. In this paper, we establish an information-theoretic framework for geo-bias evaluation, called GeoBS (Geo-Bias Scores). We demonstrate the generalizability of the proposed framework by showing how to interpret and analyze existing geo-bias measures under this framework. Then, we propose three novel geo-bias scores that explicitly take intricate spatial factors (multi-scalability, distance decay, and anisotropy) into consideration. Finally, we conduct extensive experiments on 3 tasks, 8 datasets, and 8 models to demonstrate that both task-specific GeoAI models and general-purpose foundation models may suffer from various types of geo-bias. This framework will not only advance the technical understanding of geographic bias but will also establish a foundation for integrating spatial fairness into the design, deployment, and evaluation of AI systems.

Ahmed Awadallah, Yash Lara, Raghav Magazine et al.

Progress in computer use agents (CUAs) has been constrained by the absence of large and high-quality datasets that capture how humans interact with a computer. While LLMs have thrived on abundant textual data, no comparable corpus exists for CUA trajectories. To address these gaps, we introduce FaraGen, a novel synthetic data generation system for multi-step web tasks. FaraGen can propose diverse tasks from frequently used websites, generate multiple solution attempts, and filter successful trajectories using multiple verifiers. It achieves high throughput, yield, and diversity for multi-step web tasks, producing verified trajectories at approximately $1 each. We use this data to train Fara-7B, a native CUA model that perceives the computer using only screenshots, executes actions via predicted coordinates, and is small enough to run on-device. We find that Fara-7B outperforms other CUA models of comparable size on benchmarks like WebVoyager, Online-Mind2Web, and WebTailBench -- our novel benchmark that better captures under-represented web tasks in pre-existing benchmarks. Furthermore, Fara-7B is competitive with much larger frontier models, illustrating key benefits of scalable data generation systems in advancing small efficient agentic models. We are making Fara-7B open-weight on Microsoft Foundry and HuggingFace, and we are releasing WebTailBench.

Gagan Bansal, Wenyue Hua, Zezhou Huang et al. · microsoft-research

As LLM agents advance, they are increasingly mediating economic decisions, ranging from product discovery to transactions, on behalf of users. Such applications promise benefits but also raise many questions about agent accountability and value for users. Addressing these questions requires understanding how agents behave in realistic market conditions. However, previous research has largely evaluated agents in constrained settings, such as single-task marketplaces (e.g., negotiation) or structured two-agent interactions. Real-world markets are fundamentally different: they require agents to handle diverse economic activities and coordinate within large, dynamic ecosystems where multiple agents with opaque behaviors may engage in open-ended dialogues. To bridge this gap, we investigate two-sided agentic marketplaces where Assistant agents represent consumers and Service agents represent competing businesses. To study these interactions safely, we develop Magentic-Marketplace -- a simulated environment where Assistants and Services can operate. This environment enables us to study key market dynamics: the utility agents achieve, behavioral biases, vulnerability to manipulation, and how search mechanisms shape market outcomes. Our experiments show that frontier models can approach optimal welfare -- but only under ideal search conditions. Performance degrades sharply with scale, and all models exhibit severe first-proposal bias, creating 10-30x advantages for response speed over quality. These findings reveal how behaviors emerge across market conditions, informing the design of fair and efficient agentic marketplaces.

Joykirat Singh, Justin Chih-Yao Chen, Archiki Prasad et al.

Recent thinking models solve complex reasoning tasks by scaling test-time compute, but this scaling must be allocated in line with task difficulty. On one hand, short reasoning (underthinking) leads to errors on harder problems that require extended reasoning steps; but, excessively long reasoning (overthinking) can be token-inefficient, generating unnecessary steps even after reaching a correct intermediate solution. We refer to this as under-adaptivity, where the model fails to modulate its response length appropriately given problems of varying difficulty. To address under-adaptivity and strike a balance between under- and overthinking, we propose TRAAC (Think Right with Adaptive, Attentive Compression), an online post-training RL method that leverages the model's self-attention over a long reasoning trajectory to identify important steps and prune redundant ones. TRAAC also estimates difficulty and incorporates it into training rewards, thereby learning to allocate reasoning budget commensurate with example difficulty. Our approach improves accuracy, reduces reasoning steps, and enables adaptive thinking compared to base models and other RL baselines. Across a variety of tasks (AIME, AMC, GPQA-D, BBEH), TRAAC (Qwen3-4B) achieves an average absolute accuracy gain of 8.4% with a relative reduction in reasoning length of 36.8% compared to the base model, and a 7.9% accuracy gain paired with a 29.4% length drop compared to the best RL baseline. TRAAC also shows strong generalization: although our models are trained on math datasets, they show accuracy and efficiency gains on out-of-distribution non-math datasets like GPQA-D, BBEH, and OptimalThinkingBench. Our analysis further verifies that TRAAC provides fine-grained adjustments to thinking budget based on difficulty and that a combination of task-difficulty calibration and attention-based compression yields gains across diverse tasks.

Joykirat Singh, Akshay Nambi, Vibhav Vineet

Large Language Models (LLMs) have been applied to Math Word Problems (MWPs) with transformative impacts, revolutionizing how these complex problems are approached and solved in various domains including educational settings. However, the evaluation of these models often prioritizes final accuracy, overlooking the crucial aspect of reasoning capabilities. This work addresses this gap by focusing on the ability of LLMs to detect and correct reasoning mistakes. We introduce a novel dataset MWP-MISTAKE, incorporating MWPs with both correct and incorrect reasoning steps generated through rule-based methods and smaller language models. Our comprehensive benchmarking reveals significant insights into the strengths and weaknesses of state-of-the-art models, such as GPT-4o, GPT-4, GPT-3.5Turbo, and others. We highlight GPT-$o's superior performance in mistake detection and rectification and the persistent challenges faced by smaller models. Additionally, we identify issues related to data contamination and memorization, impacting the reliability of LLMs in real-world applications. Our findings emphasize the importance of rigorous evaluation of reasoning processes and propose future directions to enhance the generalization and robustness of LLMs in mathematical problem-solving.

Somnath Kumar, Vaibhav Balloli, Mercy Ranjit et al.

Large language models (LLMs) have revolutionized various domains but still struggle with non-Latin scripts and low-resource languages. This paper addresses the critical challenge of improving multilingual performance without extensive fine-tuning. We introduce a novel dynamic learning approach that optimizes prompt strategy, embedding model, and LLM per query at runtime. By adapting configurations dynamically, our method achieves significant improvements over static, best and random baselines. It operates efficiently in both offline and online settings, generalizing seamlessly across new languages and datasets. Leveraging Retrieval-Augmented Generation (RAG) with state-of-the-art multilingual embeddings, we achieve superior task performance across diverse linguistic contexts. Through systematic investigation and evaluation across 18 diverse languages using popular question-answering (QA) datasets we show our approach results in 10-15% improvements in multilingual performance over pre-trained models and 4x gains compared to fine-tuned, language-specific models.

Anurag Ghosh, Vaibhav Balloli, Akshay Nambi et al.

Real-time perception requires planned resource utilization. Computational planning in real-time perception is governed by two considerations -- accuracy and latency. There exist run-time decisions (e.g. choice of input resolution) that induce tradeoffs affecting performance on a given hardware, arising from intrinsic (content, e.g. scene clutter) and extrinsic (system, e.g. resource contention) characteristics. Earlier runtime execution frameworks employed rule-based decision algorithms and operated with a fixed algorithm latency budget to balance these concerns, which is sub-optimal and inflexible. We propose Chanakya, a learned approximate execution framework that naturally derives from the streaming perception paradigm, to automatically learn decisions induced by these tradeoffs instead. Chanakya is trained via novel rewards balancing accuracy and latency implicitly, without approximating either objectives. Chanakya simultaneously considers intrinsic and extrinsic context, and predicts decisions in a flexible manner. Chanakya, designed with low overhead in mind, outperforms state-of-the-art static and dynamic execution policies on public datasets on both server GPUs and edge devices.