PAST-Bench Benchmark Validates Hermes Self-Improvement

A new benchmark (PAST-Bench) directly tests recursive self-improvement in personal agents. A Hermes+ variant showed measurable gains, validating Hermes Agent's core persistent memory and self-improvement claims. The improvement is real but uneven across capabilities.