IBM Research Releases VAKRA Benchmark Analysis of Agent Reasoning, Tool Use, and Failure Modes

Summary

IBM Research published a detailed analysis of VAKRA, a benchmark designed to evaluate AI agents across reasoning, tool use, and common failure modes. The blog post, authored by Ankita Naik, Danish, and other IBM researchers and hosted on Hugging Face, examines how current agentic systems perform when required to chain reasoning steps with tool invocations, and where they systematically break down. The VAKRA benchmark is significant because it moves beyond simple task completion metrics to disse