Is it agentic enough? Benchmarking open models on your own tooling

Hugging Face introduces a benchmark and harness for evaluating how efficiently open models use the transformers library in agentic tasks, measuring not just correctness but also cost, latency, and token usage across different library configurations.