DynSTEER is a dynamic stage‑wise trajectory evaluation framework designed for large language model agents performing long‑horizon tasks. It segments rollouts into stages anchored by key actions, uses a path‑tolerant milestone graph to accommodate diverse valid strategies, and adaptively routes queries to multi‑tier judges while halting unrecoverable executions early. Experiments show it improves evaluation discriminability by 85.2% over native methods, separates all model pairs with statistical significance, and saves 34.51% of execution steps on failed rollouts.
arXiv:2607. 13705v1 Announce Type: new Abstract: As Large Language Models (LLMs) evolve into autonomous agents, the need for unified evaluation infrastructure becomes critical.
As Large Language Models (LLMs) evolve into autonomous agents, the need for unified evaluation infrastructure becomes critical. However, current evaluation pipelines remain highly fragmented and tightly coupled, hindering reproducibility and causing redundant engineering.
arXiv:2603. 14465v2 Announce Type: replace Abstract: While Large Language Models (LLMs) have evolved into tool-using agents, they remain brittle in long-horizon interactions.
The paper introduces CRATE, a two‑stage vision‑language model framework that evaluates mobile agents by reasoning about each step’s consequences and aggregating this evidence to assess task completion. It also presents CRATE‑S, an extension that evaluates operational safety. Experiments show CRATE and CRATE‑S outperform existing benchmarks, achieving high F1‑scores on AndroidWorld and MobileRisk datasets.
UniACE is a unified framework that standardizes the evaluation of large language model (LLM) agents by representing each benchmark as an instruction–tool–environment triplet and running models through a shared, task‑agnostic harness in isolated runtimes. It preserves native success criteria, offers an offline mode for dynamic‑resource tasks, and standardizes efficiency metrics, execution records, and failure attribution. Applying UniACE to 7 benchmarks across 24 domains and 15 models revealed significant score shifts, ranking reversals, and sensitivity to evidence representation, highlighting the impact of evaluation configuration on reported agent performance.
arXiv:2608. 07346v1 Announce Type: new Abstract: With the rapid advancement of large language models (LLMs), harnesses have become essential infrastructure for deploying agents across a wide range of domains.
arXiv:2607. 28037v1 Announce Type: new Abstract: As LLM-based agents are deployed in complex, multi-step workflows, a critical evaluation gap has emerged: most existing benchmarks judge only final outcomes, unable to distinguish reliable reasoning from lucky success or attribute failures to specific process deficiencies, hindering attribution in long-horizon tasks.
arXiv:2608. 07346v2 Announce Type: replace Abstract: With the rapid advancement of large language models (LLMs), harnesses have become essential infrastructure for deploying agents across a wide range of domains.
arXiv:2607. 00053v1 Announce Type: cross Abstract: Large language models (LLMs) embedded in multi-turn agentic harnesses are reshaping software engineering (SWE), but routing every task to a frontier model is wasteful when many issues admit cheap fixes.
Trace2Tower is a transition‑aware EigenTrace framework that transforms raw execution traces of large language model agents into a robust skill hierarchy. By abstracting step‑level interactions into canonical events and constructing a unified graph based on semantic compatibility, transition dynamics, and outcome evidence, it isolates stable, success‑aligned behavioral modes through contrastive spectral decomposition. These modes populate a dynamic skill tower of action templates, procedural routines, and overarching task strategies, which are continuously refined via verifier‑guided feedback, achieving superior performance on ALFWorld and WebShop benchmarks.
Harbor Adapters is a unified evaluation infrastructure that ports over 80 agentic benchmarks, enabling arbitrary agents to be tested across complex environments. The authors performed a large‑scale evaluation of 8 models on 54 benchmarks, using Terminus‑2 and three native harnesses, revealing detailed agent capabilities and failure modes. They also created Harbor‑Index, a curated set of 82 challenging tasks from 29 benchmarks, designed to be affordable yet comprehensive, with the best model achieving a 28.0% pass rate.
CivBench is an open‑source benchmark that evaluates language‑model agents in the long‑horizon, tool‑mediated game Civilization VI using the Model Context Protocol (MCP). Each episode lasts over 300 turns, generating thousands of tool calls across a 76‑tool action space, and includes a narration layer that translates visual game state into structured text. The study characterises agent behaviour across four model families, introducing Proactive Monitoring Rate (PMR) and RAG@10 as interface‑level metrics, and finds that agents often under‑monitor strategic state and fail to execute near‑term commitments despite tool access and explicit guidance.