Sign InOpen Brain
arXivPaperNeeds Review

Multi-Step Tool-Calling over Korean Open Public APIs: A Benchmark and a Data-Synthesis Recipe

EDGE synthesizes multi-step tool-call training data by retaining only API links verified through live execution. Its 9B model nearly matched an untuned 27B sibling on Korean public-API tasks.

arXiv · Sep 4, 2026
Open Source Open MarkdownOpen JSON
Source Summary

KOPA-Bench contains **145 real-world tasks** over Korean open public APIs. **EDGE** maps compatible tool outputs and inputs, verifies those links against live APIs, then traverses the retained graph to generate executable multi-step trajectories.

Practical Implication

For agent training, derive tool chains from observed API behavior instead of schema compatibility alone. The reported **9B model**, fine-tuned with GRPO, nearly matched the untuned **27B model** from the same family and also improved on BFCL.

Agent-Ready Context
KOPA-Bench contains **145 real-world tasks** over Korean open public APIs. **EDGE** maps compatible tool outputs and inputs, verifies those links against live APIs, then traverses the retained graph to generate executable multi-step trajectories.

For agent training, derive tool chains from observed API behavior instead of schema compatibility alone. The reported **9B model**, fine-tuned with GRPO, nearly matched the untuned **27B model** from the same family and also improved on BFCL.

The benchmark centers on Korean government APIs and on-premise open-source deployment. The abstract reports substantial gains but gives no exact KOPA-Bench or BFCL scores, so portability to other API ecosystems remains open.
Connected Context · Feed7 Judgment

This turns multi-step tool training into a graph grounded in observed API compatibility, not schemas alone, and suggests execution-verified trajectories can let a smaller tuned model approach a larger untuned relative. Against the candidates, it reinforces deterministic validation around tool use while narrowing the evidence to Korean public APIs, one model family, and results reported without exact scores or cross-ecosystem portability.

Why Off-the-Shelf AI Doesn't Understand Money — Udi Menkes, IntuitBoth prioritize observed state-action evidence over descriptive context alone: Intuit learns from verified outcomes, while EDGE verifies tool links against live API behavior before generating trajectories.TurnSight: Turn-Level Hindsight Self-Distillation for Tool-Integrated ReasoningEDGE supplies executable multi-step trajectories, while TurnSight offers turn-level credit assignment from execution hindsight; together they address data construction and supervision granularity for tool-agent training.Twin: Playing an Unknown Game with a Test-Time Digital TwinTwin continuously validates modeled transitions before acting, reinforcing KOPA-Bench’s premise that executable behavior is stronger evidence than nominal schema compatibility, though Twin’s replayable game setting is much narrower.Healthcare’s Agent Bytecode: X12 as the Harness for AI Agents — Vasant Kearney, OnlayX12 shows how typed contracts constrain cross-system actions, while KOPA-Bench demonstrates that type compatibility still needs validation against live behavior; protocol conformance alone does not establish executable tool chains.
Context Map
agentdata#tool-use#harness-engineering#open-models
Uncertainty
The benchmark centers on Korean government APIs and on-premise open-source deployment. The abstract reports substantial gains but gives no exact KOPA-Bench or BFCL scores, so portability to other API ecosystems remains open.