aiqbench analysis
This research addresses the critical gap in multi-step tool-calling capabilities for open-source models operating within data-sovereign environments, specifically leveraging Korean government APIs. By introducing KOPA-Bench and the EDGE synthesis method, the authors demonstrate a viable path to narrowing performance disparities between smaller fine-tuned models and larger untuned counterparts without relying on proprietary infrastructure. The work provides a concrete framework for building robust, executable agent chains that adhere to strict regulatory constraints while maintaining high utility.
Multi-Step Tool-Calling over Korean Open Public APIs: A Benchmark and a Data-Synthesis Recipe
arXiv:2609.05395v1 Announce Type: cross Abstract: Data-sovereignty regulations increasingly require public institutions to deploy open-source, on-premise LLM agents that chain multiple tool-calls across live government APIs. However, open-source models consistently underperform in this multi-step setting, and no existing benchmark measures the gap. We introduce the Korean Open Public API Benchmark (KOPA-Bench), comprising 145 real-world tasks. To close this gap, we present EDGE, an Execution-grounded Dynamic Graph for tool-calling data synthEsis driven by live execution. EDGE builds a graph of how each tool's output can feed another's input, keeps only the links that succeed when actually called against the live APIs, and traverses these verified links to synthesize executable multi-step trajectories. Fine-tuned via GRPO on the resulting dataset, our 9B model nearly matches the untuned 27B model from the same family, improving substantially not only on KOPA-Bench but also on the BFCL benchmark.
Abstract:Data-sovereignty regulations increasingly require public institutions to deploy open-source, on-premise LLM agents that chain multiple tool-calls across live government APIs. However, open-source models consistently underperform in this multi-step setting, and no existing benchmark measures the gap. We introduce the Korean Open Public API Benchmark (KOPA-Bench), comprising 145 real-world tasks. To close this gap, we present EDGE, an Execution-grounded Dynamic Graph for tool-calling data synthEsis driven by live execution. EDGE builds a graph of how each tool's output can feed another's input, keeps only the links that succeed when actually called against the live APIs, and traverses these verified links to synthesize executable multi-step trajectories. Fine-tuned via GRPO on the resulting dataset, our 9B model nearly matches the untuned 27B model from the same family, improving substantially not only on KOPA-Bench but also on the BFCL benchmark.
30 pages, 7 figures, 26 tables. Accepted to EMNLP 2026 Industry Track
Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2609.05395 [cs.AI]
(or arXiv:2609.05395v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.05395 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 246 words
