Beyond Seeing: Evaluating Multimodal LLMs on Tool-Enabled Image Perception, Transformation, and Reasoning

Multimodal Large Language Models (MLLMs) are increasingly applied in real-world scenarios where user-provided images are often imperfect, requiring active image manipulations such as cropping, editing, or enhancement to uncover salient visual cues. Beyond static visual perception, MLLMs must also think with images: dynamically transforming visual content and integrating it with other tools to solve complex tasks. However, this shift from treating vision as passive context to a manipulable cognitive workspace remains underexplored. Most existing benchmarks still follow a think about images paradigm, where images are regarded as static inputs. To address this gap, we introduce IRIS, an Interactive Reasoning with Images and Systems that evaluates MLLMs' ability to perceive, transform, and reason across complex visual-textual tasks under the think with images paradigm. IRIS comprises 1,204 challenging, open-ended vision tasks (603 single-turn, 601 multi-turn) spanning across five diverse domains, each paired with detailed rubrics to enable systematic evaluation. Our evaluation shows that current MLLMs struggle with tasks requiring effective integration of vision and general-purpose tools. Even the strongest model (GPT-5-think) reaches only 18.68% pass rate. We further observe divergent tool-use behaviors, with OpenAI models benefiting from diverse image manipulations while Gemini-2.5-pro shows no improvement. By introducing the first benchmark centered on think with images, IRIS offers critical insights for advancing visual intelligence in MLLMs.

Online Rubrics Elicitation from Pairwise Comparisons

Rubrics provide a flexible way to train LLMs on open-ended long-form answers where verifiable rewards are not applicable and human preferences provide coarse signals. Prior work shows that reinforcement learning with rubric-based rewards leads to consistent gains in LLM post-training. Most existing approaches rely on rubrics that remain static over the course of training. Such static rubrics, however, are vulnerable to reward-hacking type behaviors and fail to capture emergent desiderata that arise during training. We introduce Online Rubrics Elicitation (OnlineRubrics), a method that dynamically curates evaluation criteria in an online manner through pairwise comparisons of responses from current and reference policies. This online process enables continuous identification and mitigation of errors as training proceeds. Empirically, this approach yields consistent improvements of up to 8% over training exclusively with static rubrics across AlpacaEval, GPQA, ArenaHard as well as the validation sets of expert questions and rubrics. We qualitatively analyze the elicited criteria and identify prominent themes such as transparency, practicality, organization, and reasoning.

Auto-GPT for Online Decision Making: Benchmarks and Additional Opinions

Auto-GPT is an autonomous agent that leverages recent advancements in adapting Large Language Models (LLMs) for decision-making tasks. While there has been a growing interest in Auto-GPT stypled agents, questions remain regarding the effectiveness and flexibility of Auto-GPT in solving real-world decision-making tasks. Its limited capability for real-world engagement and the absence of benchmarks contribute to these uncertainties. In this paper, we present a comprehensive benchmark study of Auto-GPT styled agents in decision-making tasks that simulate real-world scenarios. Our aim is to gain deeper insights into this problem and understand the adaptability of GPT-based agents. We compare the performance of popular LLMs such as GPT-4, GPT-3.5, Claude, and Vicuna in Auto-GPT styled decision-making tasks. Furthermore, we introduce the Additional Opinions algorithm, an easy and effective method that incorporates supervised/imitation-based learners into the Auto-GPT scheme. This approach enables lightweight supervised learning without requiring fine-tuning of the foundational LLMs. We demonstrate through careful baseline comparisons and ablation studies that the Additional Opinions algorithm significantly enhances performance in online decision-making benchmarks, including WebShop and ALFWorld.

HierCat: Hierarchical Query Categorization from Weakly Supervised Data at Facebook Marketplace

Query categorization at customer-to-customer e-commerce platforms like Facebook Marketplace is challenging due to the vagueness of search intent, noise in real-world data, and imbalanced training data across languages. Its deployment also needs to consider challenges in scalability and downstream integration in order to translate modeling advances into better search result relevance. In this paper we present HierCat, the query categorization system at Facebook Marketplace. HierCat addresses these challenges by leveraging multi-task pre-training of dual-encoder architectures with a hierarchical inference step to effectively learn from weakly supervised training data mined from searcher engagement. We show that HierCat not only outperforms popular methods in offline experiments, but also leads to 1.4% improvement in NDCG and 4.3% increase in searcher engagement at Facebook Marketplace Search in online A/B testing.

Que2Engage: Embedding-based Retrieval for Relevant and Engaging Products at Facebook Marketplace

Embedding-based Retrieval (EBR) in e-commerce search is a powerful search retrieval technique to address semantic matches between search queries and products. However, commercial search engines like Facebook Marketplace Search are complex multi-stage systems optimized for multiple business objectives. At Facebook Marketplace, search retrieval focuses on matching search queries with relevant products, while search ranking puts more emphasis on contextual signals to up-rank the more engaging products. As a result, the end-to-end searcher experience is a function of both relevance and engagement, and the interaction between different stages of the system. This presents challenges to EBR systems in order to optimize for better searcher experiences. In this paper we presents Que2Engage, a search EBR system built towards bridging the gap between retrieval and ranking for end-to-end optimizations. Que2Engage takes a multimodal & multitask approach to infuse contextual information into the retrieval stage and to balance different business objectives. We show the effectiveness of our approach via a multitask evaluation framework and thorough baseline comparisons and ablation studies. Que2Engage is deployed on Facebook Marketplace Search and shows significant improvements in searcher engagement in two weeks of A/B testing.

A Social Search Model for Large Scale Social Networks

With the rise of social networks, information on the internet is no longer solely organized by web pages. Rather, content is generated and shared among users and organized around their social relations on social networks. This presents new challenges to information retrieval systems. On a social network search system, the generation of result sets not only needs to consider keyword matches, like a traditional web search engine does, but it also needs to take into account the searcher's social connections and the content's visibility settings. Besides, search ranking should be able to handle both textual relevance and the rich social interaction signals from the social network. In this paper, we present our solution to these two challenges by first introducing a social retrieval mechanism, and then investigate novel deep neural networks for the ranking problem. The retrieval system treats social connections as indexing terms, and generates meaningful results sets by biasing towards close social connections in a constrained optimization fashion. The result set is then ranked by a deep neural network that handles textual and social relevance in a two-tower approach, in which personalization and textual relevance are addressed jointly. The retrieval mechanism is deployed on Facebook and is helping billions of users finding postings from their connections efficiently. Based on the postings being retrieved, we evaluate our two-tower neutral network, and examine the importance of personalization and textual signals in the ranking problem.