Large Language Models (LLMs) have shown impressive performance on existing medical question-answering benchmarks. This high performance makes it increasingly difficult to meaningfully evaluate and differentiate advanced methods. We present MedAgentsBench, a benchmark that focuses on challenging medical questions requiring multi-step clinical reasoning, diagnosis formulation, and treatment planning-scenarios where current models still struggle despite their strong performance on standard tests. Drawing from seven established medical datasets, our benchmark addresses three key limitations in existing evaluations: (1) the prevalence of straightforward questions where even base models achieve high performance, (2) inconsistent sampling and evaluation protocols across studies, and (3) lack of systematic analysis of the interplay between performance, cost, and inference time. Through experiments with various base models and reasoning methods, we demonstrate that the latest thinking models, DeepSeek R1 and OpenAI o3, exhibit exceptional performance in complex medical reasoning tasks. Additionally, advanced search-based agent methods offer promising performance-to-cost ratios compared to traditional approaches. Our analysis reveals substantial performance gaps between model families on complex questions and identifies optimal model selections for different computational constraints. Our benchmark and evaluation framework are publicly available at https://github.com/gersteinlab/medagents-benchmark.

Inverse protein folding, which aims to design amino acid sequences for desired protein structures, is fundamental to protein engineering and therapeutic development. While recent deep-learning approaches have made remarkable progress in addressing this challenge, they typically represent biochemical properties as discrete features associated with individual residues. Here, we present BC-Design, an approach that explicitly represents these properties as decorations on randomly sampled points on exterior surfaces and within internally bound regions representing the complete molecular extent of the protein. This provides a more natural way to capture the spatial distribution of properties. We demonstrate that BC-Design significantly outperforms all current methods, improving sequence recovery from 67% to 88.37% over the state-of-the-art methods (a 21.32% absolute improvement) and reducing perplexity from 2.4 to 1.47 (a 39.51% relative improvement) on the CATH 4.2 benchmark. Notably, our model exhibits robust generalization across diverse protein characteristics, achieving consistently high performance on proteins of varying sizes (50-500 residues), structural complexity (measured by contact order), and all major CATH fold classes. Through ablation tests, we compare the relative contribution of both structure encoding information and the encoded property information, and we show that both substantially contribute equally to this strong performance. Overall, this opens new avenues for computational protein engineering and drug discovery.

Code localization--identifying precisely where in a codebase changes need to be made--is a fundamental yet challenging task in software maintenance. Existing approaches struggle to efficiently navigate complex codebases when identifying relevant code sections. The challenge lies in bridging natural language problem descriptions with the appropriate code elements, often requiring reasoning across hierarchical structures and multiple dependencies. We introduce LocAgent, a framework that addresses code localization through graph-based representation. By parsing codebases into directed heterogeneous graphs, LocAgent creates a lightweight representation that captures code structures (files, classes, functions) and their dependencies (imports, invocations, inheritance), enabling LLM agents to effectively search and locate relevant entities through powerful multi-hop reasoning. Experimental results on real-world benchmarks demonstrate that our approach significantly enhances accuracy in code localization. Notably, our method with the fine-tuned Qwen-2.5-Coder-Instruct-32B model achieves comparable results to SOTA proprietary models at greatly reduced cost (approximately 86% reduction), reaching up to 92.7% accuracy on file-level localization while improving downstream GitHub issue resolution success rates by 12% for multiple attempts (Pass@10).

Proteins play crucial roles in biological processes, with therapeutic peptides emerging as promising pharmaceutical agents. They allow new possibilities to leverage target binding sites that were previously undruggable. While deep learning (DL) has advanced peptide discovery, generating D-proteins composed of D-amino acids remains challenging due to the scarcity of natural examples. This paper proposes D-Flow, a full-atom flow-based framework for {de novo} D-peptide design. D-Flow is conditioned on receptor binding and utilizes a comprehensive representation of peptide structure, incorporating backbone frames, side-chain angles, and discrete amino acid types. A mirror-image algorithm is implemented to address the lack of training data for D-proteins, which converts L-receptors' chirality. Furthermore, we enhance D-Flow's capacity by integrating large protein language models (PLMs) with structural awareness through a lightweight structural adapter. A two-stage training pipeline and a controlling toolkit also enable D-Flow to transition from general protein design to targeted binder design while preserving pretraining knowledge. Extensive experimental results on the PepMerge benchmark demonstrate D-Flow's effectiveness, particularly in developing peptides with entire D-residues. This approach represents a significant advancement in computational D-peptide design, offering unique opportunities for bioorthogonal and stable molecular tools and diagnostics. The code is available in~\url{https://github.com/smiles724/PeptideDesign}.

Despite Large Language Models (LLMs) like GPT-4 achieving impressive results in function-level code generation, they struggle with repository-scale code understanding (e.g., coming up with the right arguments for calling routines), requiring a deeper comprehension of complex file interactions. Also, recently, people have developed LLM agents that attempt to interact with repository code (e.g., compiling and evaluating its execution), prompting the need to evaluate their performance. These gaps have motivated our development of ML-Bench, a benchmark rooted in real-world programming applications that leverage existing code repositories to perform tasks. Addressing the need for LLMs to interpret long code contexts and translate instructions into precise, executable scripts, ML-Bench encompasses annotated 9,641 examples across 18 GitHub repositories, challenging LLMs to accommodate user-specified arguments and documentation intricacies effectively. To evaluate both LLMs and AI agents, two setups are employed: ML-LLM-Bench for assessing LLMs' text-to-code conversion within a predefined deployment environment, and ML-Agent-Bench for testing autonomous agents in an end-to-end task execution within a Linux sandbox environment. Our findings indicate that while GPT-4o leads with a Pass@5 rate surpassing 50%, there remains significant scope for improvement, highlighted by issues such as hallucinated outputs and difficulties with bash script generation. Notably, in the more demanding ML-Agent-Bench, GPT-4o achieves a 76.47% success rate, reflecting the efficacy of iterative action and feedback in complex task resolution.

Chemical reasoning usually involves complex, multi-step processes that demand precise calculations, where even minor errors can lead to cascading failures. Furthermore, large language models (LLMs) encounter difficulties handling domain-specific formulas, executing reasoning steps accurately, and integrating code effectively when tackling chemical reasoning tasks. To address these challenges, we present ChemAgent, a novel framework designed to improve the performance of LLMs through a dynamic, self-updating library. This library is developed by decomposing chemical tasks into sub-tasks and compiling these sub-tasks into a structured collection that can be referenced for future queries. Then, when presented with a new problem, ChemAgent retrieves and refines pertinent information from the library, which we call memory, facilitating effective task decomposition and the generation of solutions. Our method designs three types of memory and a library-enhanced reasoning component, enabling LLMs to improve over time through experience. Experimental results on four chemical reasoning datasets from SciBench demonstrate that ChemAgent achieves performance gains of up to 46% (GPT-4), significantly outperforming existing methods. Our findings suggest substantial potential for future applications, including tasks such as drug discovery and materials science.

The identification of protein homologs in large databases using conventional methods, such as protein sequence comparison, often misses remote homologs. Here, we offer an ultrafast, highly sensitive method, dense homolog retriever (DHR), for detecting homologs on the basis of a protein language model and dense retrieval techniques. Its dual-encoder architecture generates different embeddings for the same protein sequence and easily locates homologs by comparing these representations. Its alignment-free nature improves speed and the protein language model incorporates rich evolutionary and structural information within DHR embeddings. DHR achieves a >10% increase in sensitivity compared to previous methods and a >56% increase in sensitivity at the superfamily level for samples that are challenging to identify using alignment-based approaches. It is up to 22 times faster than traditional methods such as PSI-BLAST and DIAMOND and up to 28,700 times faster than HMMER. The new remote homologs exclusively found by DHR are useful for revealing connections between well-characterized proteins and improving our knowledge of protein evolution, structure and function.

Recently, large language models (LLMs) have evolved into interactive agents, proficient in planning, tool use, and task execution across various tasks. However, without agent-tuning, open-source models like LLaMA2 currently struggle to match the efficiency of larger models such as GPT-4 in scientific applications due to a lack of agent-tuning datasets. In response, we introduce MIMIR, a streamlined platform offering a customizable pipeline that enables users to leverage both private knowledge and publicly available, legally compliant datasets at scale for agent tuning. Additionally, MIMIR supports the generation of general instruction-tuning datasets from the same input. This dual capability ensures LLM agents developed through the platform possess specific agent abilities and general competencies. MIMIR integrates these features into an end-to-end platform, facilitating everything from the uploading of scientific data to one-click agent fine-tuning. MIMIR is publicly released and actively maintained at https://github.com/gersteinlab/MIMIR, along with a demo video1 for a quick start, calling for broader development.

Large language models (LLMs) excel at a variety of natural language processing tasks, yet they struggle to generate personalized content for individuals, particularly in real-world scenarios like scientific writing. Addressing this challenge, we introduce Step-Back Profiling to personalize LLMs by distilling user history into concise profiles, including essential traits and preferences of users. Regarding our experiments, we construct a Personalized Scientific Writing (PSW) dataset to study multiuser personalization. PSW requires the models to write scientific papers given specialized author groups with diverse academic backgrounds. As for the results, we demonstrate the effectiveness of capturing user characteristics via Step-Back Profiling for collaborative writing. Moreover, our approach outperforms the baselines by up to 3.6 points on the general personalization benchmark (LaMP), including 7 personalization LLM tasks. Our extensive ablation studies validate the contributions of different components in our method and provide insights into our task definition.

Artificial intelligence (AI)-driven methods can vastly improve the historically costly drug design process, with various generative models already in widespread use. Generative models for de novo drug design, in particular, focus on the creation of novel biological compounds entirely from scratch, representing a promising future direction. Rapid development in the field, combined with the inherent complexity of the drug design process, creates a difficult landscape for new researchers to enter. In this survey, we organize de novo drug design into two overarching themes: small molecule and protein generation. Within each theme, we identify a variety of subtasks and applications, highlighting important datasets, benchmarks, and model architectures and comparing the performance of top models. We take a broad approach to AI-driven drug design, allowing for both micro-level comparisons of various methods within each subtask and macro-level observations across different fields. We discuss parallel challenges and approaches between the two applications and highlight future directions for AI-driven de novo drug design as a whole. An organized repository of all covered sources is available at https://github.com/gersteinlab/GenAI4Drug.

The present paradigm of deep learning models for molecular representation relies mostly on 1D or 2D formats, neglecting significant 3D structural information that offers valuable physical insight. This narrow focus inhibits the model's versatility and adaptability across a wide range of modalities. Conversely, the smaller amount of research that focuses on explicit 3D representation tends to overlook textual data within the biomedical domain. We present a unified pre-trained language model that concurrently captures biomedical text, 2D, and 3D molecular information. Our model (the three-modality molecular language model, MolLM) consists of a text Transformer encoder and a molecular Transformer encoder, which encodes both 2D and 3D molecular structures. Our approach employs contrastive learning as a supervisory signal for cross-modal information learning, and we assemble a multimodality dataset using cheminformatics-based molecular modifications and a wealth of chemical text. MolLM demonstrates robust molecular representation capabilities in numerous downstream tasks, including cross-modality molecule and text matching, property prediction, captioning, and text-prompted editing. Through ablating the 3D functionality of our model, we demonstrate that the inclusion of text, 2D, and 3D representations significantly improves performance on the downstream tasks. Our code, data, and pre-trained model weights are all available at https://github.com/gersteinlab/MolLM.

Pre-trained language models like ChatGPT have significantly improved code generation. As these models scale up, there is an increasing need for the output to handle more intricate tasks. Moreover, in bioinformatics, generating functional programs poses additional notable challenges due to the amount of domain knowledge, the need for complicated data operations, and intricate functional dependencies between the operations. Here, we present BioCoder, a benchmark developed to evaluate existing pre-trained models in generating bioinformatics code. In relation to function-code generation, BioCoder covers potential package dependencies, class declarations, and global variables. It incorporates 1026 functions and 1243 methods in Python and Java from GitHub and 253 examples from the Rosalind Project. BioCoder incorporates a fuzz-testing framework for evaluation, and we have applied it to evaluate many models including InCoder, CodeGen, CodeGen2, SantaCoder, StarCoder, StarCoder+, InstructCodeT5+, and ChatGPT. Our detailed analysis of these models emphasizes the importance of domain knowledge, pragmatic code generation, and contextual understanding. Our dataset, benchmark, Docker images, and scripts required for testing are all available at https://github.com/gersteinlab/biocoder.

Large Language Models (LLMs), despite their remarkable progress across various general domains, encounter significant barriers in medicine and healthcare. This field faces unique challenges such as domain-specific terminologies and the reasoning over specialized knowledge. To address these obstinate issues, we propose a novel Multi-disciplinary Collaboration (MC) framework for the medical domain that leverages role-playing LLM-based agents who participate in a collaborative multi-round discussion, thereby enhancing LLM proficiency and reasoning capabilities. This training-free and interpretable framework encompasses five critical steps: gathering domain experts, proposing individual analyses, summarising these analyses into a report, iterating over discussions until a consensus is reached, and ultimately making a decision. Our work particularly focuses on the zero-shot scenario, our results on nine data sets (MedQA, MedMCQA, PubMedQA, and six subtasks from MMLU) establish that our proposed MC framework excels at mining and harnessing the medical expertise in LLMs, as well as extending its reasoning abilities. Based on these outcomes, we further conduct a human evaluation to pinpoint and categorize common errors within our method, as well as ablation studies aimed at understanding the impact of various factors on overall performance.

Despite the remarkable capabilities of Large Language Models (LLMs) like GPT-4, producing complex, structured tabular data remains challenging. Our study assesses LLMs' proficiency in structuring tables and introduces a novel fine-tuning method, cognizant of data structures, to bolster their performance. We unveil Struc-Bench, a comprehensive benchmark featuring prominent LLMs (GPT-NeoX-20B, GPT-3.5, GPT-4, and Vicuna), which spans text tables, HTML, and LaTeX formats. Our proposed FormatCoT aids in crafting format-specific instructions from the intended outputs to populate this benchmark. Addressing the gap in task-centered evaluation, we propose two innovative metrics, P-Score (Prompting Score) and H-Score (Heuristical Score), to more accurately gauge LLM performance. Our experiments show that applying our structure-aware fine-tuning to LLaMA-7B leads to substantial performance gains, outshining its LLM counterparts across most measures. In-depth error analysis and creating an ability map across six dimensions, coverage, formatting, reasoning, comprehension, pragmatics, and hallucination, highlight areas for future enhancements and suggest forthcoming research trajectories. Our code and models can be found at https://github.com/gersteinlab/Struc-Bench.

Large language models (LLMs) have unveiled remarkable reasoning capabilities by exploiting chain-of-thought (CoT) prompting, which generates intermediate reasoning chains to serve as the rationale for deriving the answer. However, current CoT methods either simply employ general prompts such as Let's think step by step, or heavily rely on handcrafted task-specific demonstrations to attain preferable performances, thereby engendering an inescapable gap between performance and generalization. To bridge this gap, we propose Meta-CoT, a generalizable CoT prompting method in mixed-task scenarios where the type of input questions is unknown. Meta-CoT firstly categorizes the scenario based on the input question and subsequently constructs diverse demonstrations from the corresponding data pool in an automatic pattern. Meta-CoT simultaneously enjoys remarkable performances on ten public benchmark reasoning tasks and superior generalization capabilities. Notably, Meta-CoT achieves the state-of-the-art result on SVAMP (93.7%) without any additional program-aided methods. Our further experiments on five out-of-distribution datasets verify the stability and generality of Meta-CoT.

This paper presents our contribution to the MEDIQA-2023 Dialogue2Note shared task, encompassing both subtask A and subtask B. We approach the task as a dialogue summarization problem and implement two distinct pipelines: (a) a fine-tuning of a pre-trained dialogue summarization model and GPT-3, and (b) few-shot in-context learning (ICL) using a large language model, GPT-4. Both methods achieve excellent results in terms of ROUGE-1 F1, BERTScore F1 (deberta-xlarge-mnli), and BLEURT, with scores of 0.4011, 0.7058, and 0.5421, respectively. Additionally, we predict the associated section headers using RoBERTa and SciBERT based classification models. Our team ranked fourth among all teams, while each team is allowed to submit three runs as part of their submission. We also utilize expert annotations to demonstrate that the notes generated through the ICL GPT-4 are better than all other baselines.

Software is one of the most powerful tools that we humans have at our disposal; it allows a skilled programmer to interact with the world in complex and profound ways. At the same time, thanks to improvements in large language models (LLMs), there has also been a rapid development in AI agents that interact with and effect change in their surrounding environments. In this paper, we introduce OpenHands, a platform for the development of powerful and flexible AI agents that interact with the world in similar ways to a human developer: by writing code, interacting with a command line, and browsing the web. We describe how the platform allows for the implementation of new agents, utilization of various LLMs, safe interaction with sandboxed environments for code execution, and incorporation of evaluation benchmarks. Based on our currently incorporated benchmarks, we perform an evaluation of agents over 13 challenging tasks, including software engineering (e.g., SWE-Bench) and web browsing (e.g., WebArena), amongst others. Released under the permissive MIT license, OpenHands is a community project spanning academia and industry with more than 2K contributions from over 186 contributors in less than six months of development, and will improve going forward.

We introduce MMVU, a comprehensive expert-level, multi-discipline benchmark for evaluating foundation models in video understanding. MMVU includes 3,000 expert-annotated questions spanning 27 subjects across four core disciplines: Science, Healthcare, Humanities & Social Sciences, and Engineering. Compared to prior benchmarks, MMVU features three key advancements. First, it challenges models to apply domain-specific knowledge and perform expert-level reasoning to analyze specialized-domain videos, moving beyond the basic visual perception typically assessed in current video benchmarks. Second, each example is annotated by human experts from scratch. We implement strict data quality controls to ensure the high quality of the dataset. Finally, each example is enriched with expert-annotated reasoning rationals and relevant domain knowledge, facilitating in-depth analysis. We conduct an extensive evaluation of 32 frontier multimodal foundation models on MMVU. The latest System-2-capable models, o1 and Gemini 2.0 Flash Thinking, achieve the highest performance among the tested models. However, they still fall short of matching human expertise. Through in-depth error analyses and case studies, we offer actionable insights for future advancements in expert-level, knowledge-intensive video understanding for specialized domains.

Large language models (LLMs) have dramatically enhanced the field of language intelligence, as demonstrably evidenced by their formidable empirical performance across a spectrum of complex reasoning tasks. Additionally, theoretical proofs have illuminated their emergent reasoning capabilities, providing a compelling showcase of their advanced cognitive abilities in linguistic contexts. Critical to their remarkable efficacy in handling complex reasoning tasks, LLMs leverage the intriguing chain-of-thought (CoT) reasoning techniques, obliging them to formulate intermediate steps en route to deriving an answer. The CoT reasoning approach has not only exhibited proficiency in amplifying reasoning performance but also in enhancing interpretability, controllability, and flexibility. In light of these merits, recent research endeavors have extended CoT reasoning methodologies to nurture the development of autonomous language agents, which adeptly adhere to language instructions and execute actions within varied environments. This survey paper orchestrates a thorough discourse, penetrating vital research dimensions, encompassing: (i) the foundational mechanics of CoT techniques, with a focus on elucidating the circumstances and justification behind its efficacy; (ii) the paradigm shift in CoT; and (iii) the burgeoning of language agents fortified by CoT approaches. Prospective research avenues envelop explorations into generalization, efficiency, customization, scaling, and safety. This paper caters to a wide audience, including beginners seeking comprehensive knowledge of CoT reasoning and language agents, as well as experienced researchers interested in foundational mechanics and engaging in cutting-edge discussions on these topics.

This survey addresses the crucial issue of factuality in Large Language Models (LLMs). As LLMs find applications across diverse domains, the reliability and accuracy of their outputs become vital. We define the Factuality Issue as the probability of LLMs to produce content inconsistent with established facts. We first delve into the implications of these inaccuracies, highlighting the potential consequences and challenges posed by factual errors in LLM outputs. Subsequently, we analyze the mechanisms through which LLMs store and process facts, seeking the primary causes of factual errors. Our discussion then transitions to methodologies for evaluating LLM factuality, emphasizing key metrics, benchmarks, and studies. We further explore strategies for enhancing LLM factuality, including approaches tailored for specific domains. We focus two primary LLM configurations standalone LLMs and Retrieval-Augmented LLMs that utilizes external data, we detail their unique challenges and potential enhancements. Our survey offers a structured guide for researchers aiming to fortify the factual reliability of LLMs.

Finetuning large language models (LLMs) on instructions leads to vast performance improvements on natural language tasks. We apply instruction tuning using code, leveraging the natural structure of Git commits, which pair code changes with human instructions. We compile CommitPack: 4 terabytes of Git commits across 350 programming languages. We benchmark CommitPack against other natural and synthetic code instructions (xP3x, Self-Instruct, OASST) on the 16B parameter StarCoder model, and achieve state-of-the-art performance among models not trained on OpenAI outputs, on the HumanEval Python benchmark (46.2% pass@1). We further introduce HumanEvalPack, expanding the HumanEval benchmark to a total of 3 coding tasks (Code Repair, Code Explanation, Code Synthesis) across 6 languages (Python, JavaScript, Java, Go, C++, Rust). Our models, OctoCoder and OctoGeeX, achieve the best performance across HumanEvalPack among all permissive models, demonstrating CommitPack's benefits in generalizing to a wider set of languages and natural coding tasks. Code, models and data are freely available at https://github.com/bigcode-project/octopack.

Despite the advancements of open-source large language models (LLMs), e.g., LLaMA, they remain significantly limited in tool-use capabilities, i.e., using external tools (APIs) to fulfill human instructions. The reason is that current instruction tuning largely focuses on basic language tasks but ignores the tool-use domain. This is in contrast to the excellent tool-use capabilities of state-of-the-art (SOTA) closed-source LLMs, e.g., ChatGPT. To bridge this gap, we introduce ToolLLM, a general tool-use framework encompassing data construction, model training, and evaluation. We first present ToolBench, an instruction-tuning dataset for tool use, which is constructed automatically using ChatGPT. Specifically, the construction can be divided into three stages: (i) API collection: we collect 16,464 real-world RESTful APIs spanning 49 categories from RapidAPI Hub; (ii) instruction generation: we prompt ChatGPT to generate diverse instructions involving these APIs, covering both single-tool and multi-tool scenarios; (iii) solution path annotation: we use ChatGPT to search for a valid solution path (chain of API calls) for each instruction. To enhance the reasoning capabilities of LLMs, we develop a novel depth-first search-based decision tree algorithm. It enables LLMs to evaluate multiple reasoning traces and expand the search space. Moreover, to evaluate the tool-use capabilities of LLMs, we develop an automatic evaluator: ToolEval. Based on ToolBench, we fine-tune LLaMA to obtain an LLM ToolLLaMA, and equip it with a neural API retriever to recommend appropriate APIs for each instruction. Experiments show that ToolLLaMA demonstrates a remarkable ability to execute complex instructions and generalize to unseen APIs, and exhibits comparable performance to ChatGPT. Our ToolLLaMA also demonstrates strong zero-shot generalization ability in an out-of-distribution tool-use dataset: APIBench.

We introduce FinDVer, a comprehensive benchmark specifically designed to evaluate the explainable claim verification capabilities of LLMs in the context of understanding and analyzing long, hybrid-content financial documents. FinDVer contains 4,000 expert-annotated examples across four subsets, each focusing on a type of scenario that frequently arises in real-world financial domains. We assess a broad spectrum of 25 LLMs under long-context and RAG settings. Our results show that even the current best-performing system (i.e., GPT-4o) significantly lags behind human experts. Our detailed findings and insights highlight the strengths and limitations of existing LLMs in this new task. We believe FinDVer can serve as a valuable benchmark for evaluating LLM capabilities in claim verification over complex, expert-domain documents.

Data contamination has garnered increased attention in the era of Large language models (LLMs) due to the reliance on extensive internet-derived training corpora. The issue of training corpus overlap with evaluation benchmarks—referred to as contamination—has been the focus of significant recent research. This body of work aims to identify contamination, understand its impacts, and explore mitigation strategies from diverse perspectives. However, comprehensive studies that provide a clear pathway from foundational concepts to advanced insights are lacking in this nascent field. Therefore, we present the first survey in the field of data contamination. We begin by examining the effects of data contamination across various stages and forms. We then provide a detailed analysis of current contamination detection methods, categorizing them to highlight their focus, assumptions, strengths, and limitations. We also discuss mitigation strategies, offering a clear guide for future research. This survey serves as a succinct overview of the most recent advancements in data contamination research, providing a straightforward guide for the benefit of future research endeavors.

Multimodal Large Language Models (MLLMs) have seen growing adoption across various scientific disciplines. These advancements encourage the investigation of molecule-text modeling within synthetic chemistry, a field dedicated to designing and conducting chemical reactions to synthesize new compounds with desired properties and applications. Current approaches, however, often neglect the critical role of multi-molecule graph interaction in understanding chemical reactions, leading to suboptimal performance in synthetic chemistry tasks. This study introduces PRESTO (Progressive Pretraining Enhances Synthetic Chemistry Outcomes), a new framework that bridges the molecule-text modality gap by integrating a comprehensive benchmark of pretraining strategies and dataset configurations. It progressively improves multimodal LLMs through cross-modal alignment and multi-graph understanding. Our extensive experiments demonstrate that PRESTO offers competitive results in downstream synthetic chemistry tasks. The code can be found at https://github.com/IDEA-XL/PRESTO.