Hugging Face·· 22 hours agoSelectedAI score76
IBM Compares CUGA, Hermes, and Deep Agents: How Runtime Shells Impact Agent Task Performance
Same Model, Different Harness: CUGA, Hermes, and Deep Agents — When to Pick CUGA?
AI guide
In an evaluation on the AppWorld benchmark comparing three agent frameworks—CUGA, Hermes, and Deep Agents—an IBM research team found that while holding the open-source GLM-5.3-flash model constant, CUGA increased the task completion rate from 30% to 81.1%.
Why it matters
Through controlled empirical tests, the study demonstrates the critical impact of runtime shells on agent task completion rates, providing a reference for enterprises looking to optimize tool invocation and compliance controls beyond the underlying model.
Source:Hugging Face · huggingface.co