Question 1

What is the Hermes Agent Benchmark?

Accepted Answer

Hermes Agent Benchmark is Armalo’s 16-dimension adversarial trust scorecard for AI agents. It scores agents across accuracy, reliability, self-audit (Metacal), scope honesty, safety, security, cost efficiency, latency, model compliance, runtime compliance, harness stability, and bond posture.

Question 2

How is Hermes different from a standard accuracy benchmark?

Accepted Answer

Accuracy benchmarks measure how often the agent gives the right answer. Hermes measures whether the agent can be trusted in production — including how it behaves under adversarial pressure, whether it stays within its declared scope, and whether its self-reported confidence matches reality.

Question 3

Can I run Hermes against my own agent?

Accepted Answer

Yes. Sign up for free, register your agent endpoint, and run the full battery — adversarial evals, jury arbitration, signed scorecard. The free tier covers one agent and three runs per month.

Question 4

How long does a Hermes run take?

Accepted Answer

A first-pass Hermes run completes in 5–15 minutes for most agents. The full adversarial battery with jury arbitration is run continuously after that — your score updates as your agent behaves over time.

Question 5

Is Hermes Agent Benchmark free?

Accepted Answer

The downloadable scorecard, the documentation, and the free-tier run on one agent are all free. Pro and Enterprise plans unlock continuous evals, multiple agents, audit-pack exports, and SLA-grade jury arbitration.

The 16-dimension adversarial scorecard for AI agents

Run a preview trust check on any agent

The Hermes Agent Benchmark Scorecard

Read the benchmark, chapter by chapter

The Complete Guide

Stop talking about benchmarks. Score your agent.

Architecture and Control Model

Failure Modes and Anti-Patterns

Market Map and Strategic Direction

Implementation Playbook

Security, Governance and Operational Controls

Buyer and Procurement Guide