LLM Tool Use

Large language models (LLMs) have increasingly leveraged tool invocation to enhance their reasoning capabilities. However, existing approaches typically tightly couple tool invocation with immediate execution. Such immediate tool interaction may disrupt the reasoning coherence of LLMs and constrain their expressivity, ultimately degrading reasoning performance. To this end, for the first time, we propose and formalize the problem of decoupling tool invocation from execution during reasoning, and introduce delayed execution with explicit control to enhance tool-integrated reasoning (TIR). Furthermore, we propose a hierarchical control framework and theoretically derive a surrogate loss that enables an implicitly hierarchical policy to learn behavior equivalent to that of an explicit hierarchical policy, leading to the proposed IH-GRPO algorithm. Extensive experiments on IH-GRPO achieve absolute improvements of 1.87%, 2.16%, and 2.53% on Qwen3-1.7B, Qwen3-4B, and Qwen3-8B across six out-of-domain mathematical reasoning benchmarks over the strongest baseline method, while also yielding consistent performance gains in other domains. Our code is available at https://github.com/Lumina04/IH-GRPO-01.

Large language models increasingly operate as autonomous agents that select and invoke tools from large registries. We identify a critical gap: when unauthorized tools are visible in an agent's context, models select them in adversarial scenarios -- even when explicitly instructed otherwise. We propose a governed MCP proxy that enforces attribute-based access control (ABAC) at two points: tool discovery, where unauthorized tools are removed from the model's context window, and tool invocation, where a second check blocks any unauthorized call. Across three models (Qwen 2.5 7B, Llama 3.1 8B, Claude Haiku 3.5) and 150 adversarial tasks spanning four attack categories, our proxy reduces unauthorized invocation rate (UIR) to 0% while adding under 50ms median latency. Prompt-based restrictions reduce UIR by only 11--18 percentage points, leaving substantial residual risk. Our results show that architectural enforcement -- not prompting -- is necessary for reliable tool access control in deployed agentic systems.

This research addresses the challenges inherent in developing Artificial Intelligence (AI) applications, particularly those leveraging Large Language Models (LLMs). While AI vendors provide Application Programming Interfaces (APIs) and Software Development Kits (SDKs) to facilitate developer interaction, the former often requires intricate manual request construction, and the latter can lead to significant vendor lock-in. Furthermore, existing model abstraction frameworks, though mitigating vendor dependency, introduce an additional layer of complexity and potential security concerns. To reconcile these conflicting factors, the study introduces PuppyChatter, a novel software framework designed to preserve the intuitive simplicity of vendor-specific SDKs while simultaneously adhering to the vendor-neutrality principles characteristic of model abstraction, thereby offering a more streamlined and flexible development paradigm.

Large language models are increasingly used as planning components in agentic systems, but current tool-use pipelines often require full tool schemas to be included in every prompt, creating substantial token overhead and limiting the practicality of smaller models. This paper investigates whether tool-use knowledge can be internalized into small language models through parameter-efficient fine-tuning, enabling structured planning without explicit tool descriptions at inference time. Using AssetOpsBench as the primary benchmark, we fine-tune Gemma 4 E4B and Qwen3-4B with 8-bit QLoRA on approximately 1,700 tool-use examples spanning tool knowledge, question-to-plan mappings, and execution-style traces. We evaluate the resulting models under description-free inference, where the prompt omits the tool catalog entirely. The fine-tuned models outperform an informed unfine-tuned baseline that receives full tool descriptions, reducing input length by 82.6% while improving structural and LLM-judge planning scores. In the best Gemma run, the model achieves an AT-F1 of 0.65 and an overall judge score of 3.88, compared with 0.47 and 2.88 for the informed baseline. Qwen3-4B achieves a strong overall judge score of 3.78 while using 62% less memory and running 2.5$\times$ faster than Gemma, though it also exhibits greater catastrophic forgetting on general multiple-choice benchmarks. Additional ablations show that LoRA rank controls a quality--retention trade-off, with $r=32$ maximizing planning quality and smaller ranks preserving more general knowledge. These results suggest that, for fixed tool catalogs, QLoRA fine-tuning can shift tool knowledge from prompt context into model weights, substantially reducing inference overhead while maintaining or improving tool-planning quality.

Training tool-calling agents requires large-scale trajectory data with verifiable labels, yet existing approaches either synthesize environments that diverge from real API behavior or generate tasks without ground-truth outcomes for verification. We present FireFly, a pipeline for generating verified tool-call data from real-world MCP servers. Our key insight is to invert the standard synthesis pipeline: rather than generating tasks and hoping they are solvable, we first let a strong LLM explore real APIs along graph-guided DAG structures, then synthesize tasks backward from observed outcomes, guaranteeing label correctness by construction. To handle the scale of real-world tool spaces (${\sim}$1,000 tools), we build a pairwise tool graph and sample sub-DAGs to focus exploration on semantically coherent workflows. To address environment drift in live APIs, we construct a retrieval-augmented simulator that caches all exploration results and replays them during training and evaluation, enabling fully offline and reproducible RL. Applying this pipeline yields 5,144 verified tasks spanning 240 servers and 993 tools. A 4B-parameter model trained with GRPO on FireFly matches Claude Sonnet 4.6 on our held-out test set and shows improvements on multiple tool-calling benchmarks including Tau2-Bench, MCPMark, and MCP-Atlas.

Function calling, also known as tool use, is a core capability of modern LLM agents but is typically constrained by synchronous execution semantics. Under these semantics, LLM decoding is blocked until each function call completes, resulting in increasing end-to-end latency. In this work, we introduce AsyncFC, a pure execution-layer framework that decouples LLM decoding from function execution, enabling overlap between model decoding and function execution as well as inter-function parallelism when dependencies permit. AsyncFC layers over existing models and unmodified function implementations, requiring no fine-tuning or changes to the standard synchronous function-calling protocol. Across standard function-calling benchmarks and adapted software engineering benchmarks, AsyncFC significantly reduces end-to-end task completion time while preserving task accuracy. Furthermore, these results reveal that LLMs possess a native capability to reason over symbolic futures that represent unresolved execution results, enabling an asynchronous paradigm for model-tool interaction.

Tool use extends large language models beyond parametric knowledge, but reliable execution requires balancing appropriate reasoning depth with strict structural validity. We approach this problem from a case-based perspective to present CAST, a case-driven framework that treats historical execution trajectories as structured cases. Instead of reusing raw exemplar outputs, CAST extracts case-derived signals to identify complexity profiles for estimating optimal reasoning strategies, alongside failure profiles to map likely structural breakdowns. The framework translates this knowledge into a fine-grained reward design and adaptive reasoning, enabling the model to autonomously internalize case-based strategies during reinforcement learning. Experiments on BFCLv2 and ToolBench demonstrate that CAST improves both schema-faithful execution and task-level tool-use success while reducing unnecessary deliberation. The approach achieves up to 5.85 percentage points gain in overall execution accuracy and reduces average reasoning length by 26%, significantly mitigating high-impact structural errors. Ultimately, this demonstrates how historical execution cases can provide reusable adaptation knowledge for calibrated tool use.

Tool-augmented LLM agents increasingly access the same tool type through multiple functionally equivalent providers, such as web-search APIs, retrievers, or LLM backends exposed behind a shared interface. This creates a provider-routing problem under runtime load: the router must choose among providers that differ in latency, reliability, and answer quality, often without gold labels at deployment time. We introduce LQM-ContextRoute, a contextual bandit router for same-function tool providers. Its key design is latency-quality matching: instead of letting low latency offset poor answers in an additive reward, the router ranks providers by expected answer quality per service cycle. It combines this capacity-aware score with query-specific quality estimation and LLM-as-judge feedback, allowing it to adapt online to both load changes and provider-quality differences. On the main web-search load benchmark, LQM-ContextRoute improves F1 by +2.18 pp over SW-UCB while staying on the latency-quality frontier. In a high-heterogeneity StrategyQA setting, LQM-ContextRoute avoids additive-reward collapse and improves accuracy by up to +18 pp over SW-UCB; on heterogeneous retriever pools, it improves NDCG by +2.91--+3.22 pp over SW-UCB. These results show that same-function tool routing benefits from treating latency as service capacity, especially when runtime pressure and provider-quality heterogeneity coexist.

Large language models (LLMs) increasingly act as autonomous agents that must decide when to answer directly vs. when to invoke external tools. Prior work studying adaptive tool use has largely treated tool necessity as a model-agnostic property, annotated by human or LLM judge, and mostly cover cases where the answer is obvious (e.g., fetching the weather vs. paraphrasing text). However, tool necessity in the wild is more nuanced due to the divergence of capability boundaries across models: a problem solvable by a strong model on its own may still require tools for a weaker one. In this work, we introduce a model-adaptive definition of tool-necessity, grounded in each model's empirical performance. Following this definition, we compare the necessity against observed tool-call behavior across four models on arithmetic and factual QA dataset, and find substantial mismatches of 26.5-54.0% and 30.8-41.8%, respectively. To diagnose the failure, we decompose tool use into two stages: an internal cognition stage that reflects whether a model believes a tool is necessary, and an execution stage that determines whether the model actually makes a tool-call action. By probing the LLM hidden states, we find that both signals are often linearly decodable, yet their probe directions become nearly orthogonal in the late-layer, last-token regime that drives the next-token action. By tracing the trajectory of samples in the two-stage process, we further discover that the majority of mismatch is concentrated in the cognition-to-action transition, not in cognition itself. These results reveal a knowing-doing gap in LLM tool-use: improving tool-use reliability requires not only better recognition of when tools are needed, but also better translation of that recognition into action.

We present VectraYX-Nano, a 41.95M-parameter decoder-only language model trained from scratch in Spanish for cybersecurity, with a Latin-American regional focus and native tool invocation via the Model Context Protocol (MCP). The model has four contributions. (i) Corpus: VectraYX-Sec-ES, a 170M-token Spanish corpus assembled by an eight-VM distributed pipeline at ~$25 USD of cloud compute and split into three curriculum phases (conversational 42M, cybersecurity 118M, offensive tooling 10M). (ii) Architecture: a 42M Transformer decoder with GQA, QK-Norm, RMSNorm, SwiGLU, RoPE and z-loss, paired with a domain-balanced 16,384-token byte-fallback BPE. (iii) Curriculum with replay across the three phases yields a monotonic loss descent (9.80 -> 3.17 -> 3.00 -> 2.16); after SFT (loss 1.74) the v2 bootstrap-ablation reference attains a conversational gate of 0.775 +/- 0.043 on B5 over N=4 seeds, and a controlled Phase-2 replay sweep over {0,5,10,25,50}% saturates B5 at >=25% replay. (iv) Two empirical findings, both N=4. A controlled bootstrap-corpus ablation across v2 (OpenSubs), v4 (mC4-ES), and v6 (60/25/15 OpenSubs/mC4/Wiki) exposes a loss-versus-register inversion: lower-perplexity bootstraps yield measurably worse conversational behavior (v2 > v4 > v6 on B5 at every paired seed). The B4 (tool-selection) floor of 0.000 is a corpus-density artifact, not a capacity gate: rebalancing the SFT mixture to tool-use ratio 1:21 yields VectraYX-Nano v7, the released headline configuration, reaching B4 = 0.230 +/- 0.052 at 42M while retaining B1 = 0.332 +/- 0.005 and B5 = 0.725 +/- 0.130; a LoRA replication on a 260M from-scratch mid-tier reaches 0.445 +/- 0.201. The released GGUF is 96 MB in F16, runs sub-second TTFT on commodity hardware under llama.cpp, and is, to our knowledge, the first published Spanish-native cybersecurity LLM with end-to-end MCP integration.

Language models can produce convincing scientific analyses, but repeated generations on the same data do not guarantee the same result. A researcher may regenerate an identical query and receive a different fit, a different peak position or a different analysis procedure, without an obvious way to decide which output to trust. We propose typed mediation, a pattern in which the model orchestrates deterministic tools rather than generating analytical code. Each tool encodes one researcher's exact procedure for one instrument, ported through structured interviews. The model selects which tool to call and with what parameters. The tool produces the result. Regeneration does not change it. We evaluate this claim by running the same photoluminescence analysis on four platforms, including three commercial foundation models, four times each with the same prompt. The typed tool produces identical results across all runs. The commercial platforms either vary in numerical output and analytical methodology across runs, or fail to produce valid results on the task. We deploy this pattern on two instruments serving users over approximately six months, with very positive user feedback. Both cases are very challenging: they involve proprietary binary formats and per-seat licensed software, which force the tool to remain on local infrastructure alongside the data and the instrument it operates. We argue that deployment topology is not just a preference, but a structural requirement of scientific tool mediation. The result is a practical pattern for deploying language models in scientific workflows where reproducibility is mandatory, reducing analysis time from weeks to minutes while guaranteeing identical outputs across runs.

Advances in large language models (LLMs) have recently opened new and promising avenues for small-molecule drug discovery. Yet existing LLM-based approaches for molecular generation often suffer from high rates of invalid and low-quality ligand candidates, a result of the syntactic limitations of current models with regard to molecular strings. In this paper, we introduce $\texttt{ToolMol}$, an evolutionary agentic framework for de novo drug design. $\texttt{ToolMol}$ combines a multi-objective genetic algorithm with an agentic LLM operator that iteratively updates the ligand population. We build a comprehensive toolbox of RDKit-backed functions that allows our agentic operator to consisently make precise ligand modifications. $\texttt{ToolMol}$ achieves state-of-the-art performance on multi-objective property optimization tasks, discovering drug-like and synthesizable ligands that have $>10\%$ stronger predicted binding affinity compared to existing methods, evaluated on three protein targets. $\texttt{ToolMol}$ ligands additionally achieve state-of-the-art results in gold-standard Absolute Binding Free Energy scores, gaining over existing methods by over $35\%$. By studying chain-of-thought reasoning traces, we observe that tool-calling enables the model to more faithfully execute its planned modifications, efficiently exploiting the strong chemical prior knowledge in LLMs.

The power distribution engineering workforce faces a projected shortage of up to 1.5 million engineers by 2030, creating urgent demand for more accessible analysis tools. This paper introduces Grid-Orch, a framework that bridges Large Language Models (LLMs) and power system simulation through the Model Context Protocol (MCP), enabling engineers to perform complex distribution analyses via natural language. Using OpenDSS as the reference implementation, Grid-Orch provides 36 domain-specific tools across eleven categories, covering power flow, voltage analysis, quasi-static time series (QSTS) simulation, and automated optimization. A provider-agnostic LLM layer supports both cloud-hosted (Gemini, Claude) and locally deployed (Ollama, llama-cpp) models, enabling air-gapped operation for security-sensitive utility environments. Three optimization skills, capacitor placement, voltage violation analysis, and overvoltage mitigation, extend the platform beyond single-tool queries to multi-step engineering workflows. Grid-Orch is delivered as an interactive web platform with chat-based interaction, a QSTS dashboard, and feeder topology visualization, and renders simulation results inline. Workflow demonstrations show that distribution analyses formerly requiring hours of scripting, such as distributed energy resource (DER) interconnection screening, complete in under two minutes through natural language, producing numerically identical results to direct OpenDSS scripting.

Existing multi-model and tool-augmented systems communicate by generating text, serializing every exchange through the output vocabulary. Can two pretrained language models instead coordinate through a continuous, concurrent channel? The Bicameral Model couples two frozen language models through a trainable neural interface on their intermediate hidden states. At every generation step, both models run in lockstep: a primary model drives the task while an auxiliary model operates tools, solves constraints, or executes code, with both conditioning on each other's activations through a translation network and a learned suppression gate ($\sim$1% of combined parameters). The gate learns a selective communication protocol from task loss alone, without a prescribed format. We demonstrate the mechanism across three tool backends. On arithmetic, coupling two 0.5B models with a calculator raises accuracy from 36% to 96%. On logic grid puzzles, coupling two 0.6B models with a Z3 solver achieves $1.7\times$ the unaugmented baseline on ZebraLogic. On mathematical reasoning, coupling with a Python sandbox enables the auxiliary to generate problem-specific code from hidden-state signals alone, without ever seeing the problem text.

Tool-integrated reasoning (TIR) enables large language models (LLMs) to enhance their capabilities by interacting with external tools, such as code interpreters (CI). Most recent studies focus on exploring various methods to equip LLMs with the ability to use tools. However, how to further boost the reasoning ability of already tool-capable LLMs at inference time remains underexplored. Improving reasoning at inference time requires no additional training and can help LLMs better leverage tools to solve problems. We observe that, during tool-capable LLM inference, both the number and the proportion of erroneous tool calls are negatively correlated with answer correctness. Moreover, erroneous tool calls are typically resolved successfully within a few subsequent turns. If not, LLMs often struggle to resolve such errors even with many additional turns. Building on the above observations, we propose PruneTIR, a rather effective yet efficient framework that enhances the tool-integrated reasoning at inference time. During LLM inference, PruneTIR prunes trajectories, resamples tool calls, and suspends tool usage through three components: Success-Triggered Pruning, Stuck-Triggered Pruning and Resampling, and Retry-Triggered Tool Suspension. These three components enable PruneTIR to mitigate the negative impact of erroneous tool calls and prevent LLMs from getting stuck in repeated failed resolution attempts, thereby improving overall LLM performance. Extensive experimental results demonstrate the effectiveness of PruneTIR, which significantly improves Pass@1 and efficiency while reducing the working context length for tool-capable LLMs.

Most language-model training data shows final artifacts, not the process that produced them. We study a tractable version of this question in tool use: when a model learns a stream of new API domains, does keeping tool-use trajectories help compared with stripping the intermediate API trace? We fine-tune Llama 3.1 8B Instruct with QLoRA on API-Bank using four sequential domain blocks. Condition A strips previous API request/response lines from the prompt and trains the model to predict the next API call. Condition B keeps the trajectory context. In a single-seed pilot, full held-out generation evaluation shows that Condition B reaches 56.9% final exact full-call accuracy compared with 39.2% for Condition A. B also improves final API-name accuracy by 7.7 points. However, B uses 25.1% more training tokens, the run uses one seed, and the task is next-call prediction rather than full dialogue success.

Tool-augmented LLM agents tend to call tools indiscriminately, even when the model can answer directly. Each unnecessary call wastes API fees and latency, yet no existing benchmark systematically studies when a tool call is actually needed. We propose When2Tool, a benchmark of 18 environments (15 single-hop, 3 multi-hop) spanning three categories of tool necessity -- computational scale, knowledge boundaries, and execution reliability -- each with controlled difficulty levels that create a clear decision boundary between tool-necessary and tool-unnecessary tasks. We evaluate two families of training-free baselines: Prompt-only (varying the prompt to discourage unnecessary calls) and Reason-then-Act (requiring the model to reason about tool necessity before acting). Both provide limited control: Prompt-only suppresses necessary calls alongside unnecessary ones, and Reason-then-Act still incurs a disproportionate accuracy cost on hard tasks. To understand why these baselines fail, we probe the models' hidden states and find that tool necessity is linearly decodable from the pre-generation representation with AUROC 0.89--0.96 across six models, substantially exceeding the model's own verbalized reasoning. This reveals that models already know when tools are needed, but fail to act on this knowledge during generation. Building on this finding, we propose Probe&Prefill, which uses a lightweight linear probe to read the hidden-state signal and prefills the model's response with a steering sentence. Across all models tested, Probe&Prefill reduces tool calls by 48% with only 1.7% accuracy loss, while the best baseline at comparable accuracy only reduces 6% of tool calls, or achieves a similar tool call reduction but incurs a 5$\times$ higher accuracy loss. Our code is available at https://github.com/Trustworthy-ML-Lab/when2tool

We present Open Ontologies, an open-source ontology engineering system implemented in Rust that integrates LLM-driven construction with formal OWL reasoning and ontology alignment via the Model Context Protocol. Our primary finding is that stable 1-to-1 matching is the dominant factor in ontology alignment quality: on the OAEI Anatomy track, it achieves F1 = 0.832 (P = 0.963, R = 0.733), competitive with state-of-the-art systems and exceeding all in precision. Ablation across five weight configurations shows that signal weights are irrelevant when stable matching is applied (F1 varies by less than 0.004), while removing stable matching drops F1 to 0.728. On the Conference track, the same method achieves F1 = 0.438. On tool-augmented ontology interaction, we find a surprising result: an LLM reading a raw OWL file (F1 = 0.323) performs worse than the same LLM with no file at all (F1 = 0.431), while structured MCP tool access achieves F1 = 0.717. This demonstrates that tool structure provides a qualitatively different mode of access that the LLM cannot replicate by reading raw syntax. The system ships as a single binary under the MIT licence.

Teaching language models to use search tools is not only a question of whether they search, but also of whether they issue good queries. This is especially important in open-domain question answering, where broad or copied queries often waste retrieval budget and derail later reasoning. We propose \Ours, a framework that makes query planning explicit through reusable search skills. At each step, the model first selects a skill, then generates a search or answer action conditioned on the selected skill card. The skill inventory itself is not fixed: SearchSkill maintains an evolving SkillBank, expands or refines it from recurrent failure patterns, and reconstructs affected trajectories before supervised training. The resulting two-stage SFT recipe aligns training with the inference-time protocol of skill selection followed by skill-grounded execution. Across open-source and closed-source models, SearchSkill improves exact match on knowledge-intensive QA benchmarks and yields better retrieval behavior, including fewer copied first queries, more atomic hop-focused queries, and more correct answers within a small search budget. These results suggest that explicit skill-conditioned query planning is a lightweight alternative to treating search as an undifferentiated action.

We introduce MathConstraint, a hard, adaptive benchmark for evaluating the combinatorial reasoning capabilities of LLMs. We combine constraint satisfaction problems with rigorous solver-based verification and design an adaptive generator to create instances that remain challenging as the LLMs improve in their reasoning capabilities. Unlike existing benchmarks that quickly saturate on fixed datasets or use LLM-as-a-judge for checking solutions,MathConstraint uses parameterized problem types that enable scalable generation of arbitrarily difficult and automatically verifiable instances. We release MathConstraint-Easy ($266$ instances), on which frontier models achieve between $72.6\%$ (gemini-3.1-flash-lite) and $87.6\%$ (gpt-5.5) accuracy, and MathConstraint ($329$ instances) on which the same models drop to between $18.5\%$ (claude-4.6-sonnet) and $66.9\%$ (gpt-5.5) accuracy, demonstrating the resilience of our benchmark generator against rapid progress in LLM reasoning capabilities. We evaluate 12 frontier and open-weight models with and without access to a sandboxed Python environment that includes generic SAT/SMT solvers. Tool access roughly doubles frontier accuracy on MathConstraint (mean $+28$pp; up to $+52$pp for claude-4.6-sonnet). Further, halving the tool-call budget from $8$ to $4$ rounds erases up to $37$ points -- a sensitivity that most single-budget benchmarks miss. We release the generator, dataset, and evaluation harness as a robust environment for studying combinatorial reasoning and tool-use behavior under adversarially-tunable difficulty.

Tool-augmented language models can extend small language models with external executable skills, but scaling the tool library creates a coupled challenge: the library must evolve with the planner as new reusable subroutines emerge, while retrieval from the growing library must remain within a fixed context budget. Existing tool-use and skill-library methods typically treat tools as flat or text-indexed memories, causing prompt cost to grow with library size and obscuring the typed, compositional structure of executable code. We propose CoCoDA, a framework that co-evolves the planner and tool library through a single code-native structure: a compositional code DAG. Nodes are primitive or composite tools, edges encode invocation dependencies, and each node stores a typed signature, description, pre/post-condition specification, and worked examples. At inference time, Typed DAG Retrieval prunes candidates by symbolic signature unification, ranks survivors by descriptions, filters them by behavioral specifications, and disambiguates with examples, keeping expensive context materialization on progressively smaller candidate sets. At training time, successful trajectories are folded into validated composite tools, while the planner is updated with a DAG-induced reward that credits composites by their primitive expansion size. We provide theoretical results showing retrieval cost reduction, sublinear retrieval time, compositional advantage under the shaped reward, monotone co-evolution under conservative updates, and DAG well-formedness. Across mathematical reasoning, tabular analysis, and code task benchmarks, CoCoDA enables an 8B student to match or exceed a 32B teacher on GSM8K and MATH and consistently improves over strong tool-use and library-learning baselines.

When a tool-calling agent picks the wrong tool, the failure is invisible until execution: the email gets sent, the meeting gets missed. As agents take on consequential actions, one bad tool call can do real damage. We currently have no way to look inside the model and catch the mistake before it happens; this paper shows that we can. Inside the model, the choice of tool is carried by a single direction in activation space, one direction per pair of tools. Adding that direction during generation switches which tool the model picks. Across 12 instruction-tuned and 6 base models spanning Gemma 3, Qwen 3, Qwen 2.5, and Llama 3.1 (270M to 27B), this works at 83-100% accuracy on 4B+ instruction-tuned models on a 15-tool synthetic benchmark and at 77-94% on the real-API benchmark $τ$-bench airline. The JSON arguments that follow automatically adapt to the new tool's schema, so flipping the name is enough. The same per-tool directions also flag likely errors before they happen: queries where the model is unsure between two tools fail 21x more often than queries where it is not (Gemma 3 27B). This is not just topic injection: random vectors at the same magnitude give a 0% switch rate, and a probe within a single domain (14 airline tools that share one topic) still reads which tool the model will call at top-1 61-89% across five 4B-14B models. Even base models already carry the right tool internally before they can emit it: reading the chosen tool off the model's internal state (cosine readout) recovers 61-82% accuracy on BFCL while base generation lands at 2-10%, suggesting pretraining forms the representation and instruction tuning later wires it to the output. Our results cover single-turn, fixed-menu settings; on multi-turn agent loops the same intervention is less stable (matched-baseline gain or loss of up to 30 percentage points with no consistent direction).

Large Language Models (LLMs) have become increasingly capable as tool-using agents, with benchmarks spanning diverse general agentic tasks. Yet rigorous evaluation of scientific tool use remains limited. In chemistry, recent agents can plan syntheses and invoke domain-specific tools, but evaluations often rely on curated demonstrations, expert assessment, or LLM-as-judge scoring rather than exact, judge-free ground truth. We address this gap with chemical procurement cost estimation, a practical task in which an agent must ground chemical identities, retrieve supplier quotes, select valid purchasable packs, normalize quantities, and compute cost from a reaction description. We introduce ChemCost, a benchmark of 1,427 evaluable reactions grounded to a frozen pricing snapshot covering 2,261 chemicals and 230,775 supplier quotes, supporting scalar scoring and stage-level diagnosis of grounding, retrieval, procurement, and arithmetic failures. To evaluate robustness, we further construct controlled noise-injected views that perturb chemical aliases, quantity expressions, missing fields, and input formatting. Experiments with frontier, open-weight, and chemistry-specialized LLM agents show that tool access is necessary but insufficient for solving the task. The strongest agents reach only 50.6% accuracy within 25% relative error on clean inputs and degrade substantially with realistic noise. Stage-level analysis further shows that failures arise from brittle parsing, ineffective evidence integration, invalid pack selection, and non-convergent tool use.

Agentic AI systems that invoke external tools are powerful but costly, leading developers to default to large models and overspend inference budgets. Model routing can mitigate this, but existing routers are designed for chat completion rather than tool use. We present Switchcraft, the first (to the best of our knowledge) model router optimized for agentic tool calling. Switchcraft operates inline, selecting the lowest-cost model subject to correctness. We construct an evaluation framework on five function-calling benchmarks and train a DistilBERT-based classifier, deployed under a latency budget. Switchcraft achieves 82.9% accuracy -- matching or exceeding the best individual model -- while reducing inference cost by 84%, saving over $3,600 per million queries. We find that larger models do not consistently outperform smaller ones on tool-use tasks, and that nominally cheaper models can incur higher total cost due to token-intensive reasoning. Our work enables cost-aware agentic AI deployment without sacrificing correctness.

Reaction feasibility prediction, as a fundamental problem in computational chemistry, has benefited from diverse tools enabled by recent advances in artificial intelligence, particularly large language models. However, the performance of individual tools varies substantially across reactions, making it difficult for any single tool to consistently perform well across all cases. This raises a critical challenge: how to effectively leverage multiple tools to obtain more accurate feasibility predictions. To address this, we propose ARMOR, an agentic framework that explicitly models tool-specific utilities, adaptively prioritizes tools, and further resolves the potential tool conflicts to produce the final prediction for each reaction. Unlike existing approaches that rely on simple aggregation or heuristic assignment over various tools, ARMOR organizes tools into a hierarchy that prioritizes top-performing tools and defers others when needed, characterizes their strengths through tool-specific patterns, and resolves conflicts via memoryaugmented reasoning. Extensive experiments on a public dataset demonstrate that ARMOR consistently outperforms strong baselines, including single-tool methods as well as various tool aggregation and tool selection approaches. Further analysis shows that the improvements are particularly significant on reactions with conflicting tool predictions, highlighting the effectiveness of ARMOR in leveraging the complementary strengths of multiple tools. The code is available via https://anonymous.4open.science/r/ARMOR-E13F.

This paper introduces tool efficiency, a new quantitative metric to evaluate the rate of useful tool calls in an LLM agent trajectory. To ensure that tool efficiency is well-defined, we also introduce marginal tool utility, a new quantitative metric defined per tool call indicating whether a tool is useful or whether it can be safely removed from the tool suite without affecting accuracy while increasing tool efficiency; in this paper, we determine the sign of marginal tool utility for each tool call in a trajectory using LLM-as-a-Judge. While much prior work has been done to develop techniques that improve tool use by LLMs and design evaluation methods measuring efficiency indirectly using accuracy as a proxy, our work is centered on measuring efficiency directly via the quantitative metric proposed in this paper in post hoc trajectory analyses. It is our intention that this work contributes to the frontier of LLM evaluation research as a springboard for future benchmark designs and agent harness engineering (specifically with regards to creating lean tool suites) that optimize for metrics that complement but are distinct from accuracy.

Signal Temporal Logic (STL) is an expressive formal language for specifying spatio-temporal requirements over real-valued, real-time signals. It has been widely used for the verification and synthesis of autonomous systems and cyber-physical systems. In practice, however, users often express their requirements in natural language rather than in structured STL formulas, making natural-language-to-STL translation a critical yet challenging task. Manual specification requires temporal-logic expertise and cannot scale, while prompting commercial LLM APIs incurs substantial token costs and may expose sensitive system requirements to third-party services, raising privacy concerns for industrial deployment. To address these challenges, we present \textsc{ReasonSTL}, a tool-augmented framework that adapts local open-source language models for natural-language-to-STL generation. \textsc{ReasonSTL} decomposes the translation process into explicit reasoning, deterministic tool calls, and structured formula construction. We further introduce process-rewarded training to supervise both tool-use trajectories and final formulas, together with \textsc{STL-Bench}, a bilingual, computation-aware benchmark grounded in real-world signals. Experiments show that a 4B model trained with \textsc{ReasonSTL} achieves state-of-the-art performance in both automatic metrics and human evaluations, demonstrating that \textsc{ReasonSTL} provides a transparent, low-cost, and privacy-preserving alternative for formal specification drafting.

Despite the success of large language models (LLMs) on general-purpose tasks, their performance in highly specialized domains such as biomedicine remains unsatisfactory. A key limitation is the inability of LLMs to effectively leverage biomedical tools, which clinical experts and biomedical researchers rely on extensively in daily workflows. While recent general-domain tool-calling datasets have substantially improved the capabilities of LLM agents, existing efforts in the biomedical domain largely rely on in-context learning and restrict models to a small set of tools. To address this gap, we introduce BioTool, a comprehensive biomedical tool-calling dataset designed for fine-tuning LLMs. BioTool comprises 34 frequently used tools collected from the NCBI, Ensembl, and UniProt databases, along with 7,040 high-quality, human-verified query-API call pairs spanning variation, genomics, proteomics, evolution, and general biology. Fine-tuning a 4-billion-parameter LLM on BioTool yields substantial improvements in biomedical tool-calling performance, outperforming cutting-edge commercial LLMs such as GPT-5.1. Furthermore, human expert evaluations demonstrate that integrating a BioTool-fine-tuned tool caller significantly improves downstream answer quality compared to the same LLM without tool usage, highlighting the effectiveness of BioTool in enhancing the biomedical capabilities of LLMs. The full dataset and evaluation code are available at https://github.com/gxx27/BioTool

General-purpose LLMs with web search are increasingly used to scout the competitive landscape of pharmaceutical pipelines. We benchmark Gosset -- an AI platform with a chat interface backed by curated target-, modality-, and indication-level drug-asset annotations -- against four frontier systems with web access (Claude Opus 4.7, GPT 5.5, Gemini 3.1 Pro, Perplexity sonar-pro) on ten niche oncology/immunology targets where most of the pipeline lives in the long tail of preclinical and Asian-developed assets. All five systems receive the same natural-language query and the same JSON output schema. Across 10 targets Gosset returns 3.2x more verified drugs per query than the best frontier system, at perfect precision and 100% recall against the cross-system union of verified drugs. The same curated index is exposed as a Gosset MCP server that any frontier model can call as a tool, suggesting that each of these systems can close most of the recall gap by swapping generic web search for a curated index behind the same chat interface.

There is an increasing number of algorithmic computations in theoretical physics. These, while conceptually simple, can nevertheless be time-consuming and contain subtleties that should not be overlooked. Given the recent improvement of Large Language Models (LLM), it is natural to investigate whether LLMs equipped with a computer algebra system (CAS) runtime and sufficiently informative context can reliably carry out these algorithmic tasks. In this work, we interface Claude with Maple, and apply this framework to cosmological perturbations in modified theories of gravity. We demonstrate the current capabilities of this approach, the typical failures, and how the same can be improved. We find that a frontier LLM supplied with worked examples is able to solve most test problems.