Hermes Agent For Business + Which AI Models To Run For Each Job: Real GoldieBench Scores For Bots, Boards And Client Pages

Hermes Agent for business lets you run a different AI model for each job, so the useful question isn't "which model is best" but "which model fits this Bot, this channel or this card".

Below are the GoldieBench scores I read on the live pages on 4 November 2026, matched to the places Hermes lets you pin a model.

The examples are a salon, an accountant, a shop and an agency, and each pick says what the model actually does in that job.

Want my team to map models to your own workflows? Book a free SEO strategy session.

What GoldieBench Measures, And What It Doesn't

The GoldieBench methodology gives every model the same fixed prompt, one shot, and the result must be a single self-contained HTML file.

Julian scores each result from 0 to 10, and the leaderboard ranks models by their average.

So the scores measure building things in one go: games, simulations and web pages.

They don't measure inbox triage, report writing or reading a PDF.

Use them for the build-heavy jobs, and as a rough guide to overall strength for the rest.

Five Places Hermes Lets You Pin A Model

• The main model: set with hermes model, used by anything that isn't pinned.
• Each cron job: a per-job pin, or a cron.model default for every unpinned job, per the scheduled tasks docs.
• Each Bot: Bot Mode's New Agent dialog has a model and provider pin, and different Bots can run different models side by side.
• Each chat channel: channel_overrides in the messaging gateway set a model and system prompt per channel.
• Each Kanban card: --model at creation or hermes kanban set-model later.

That's why one install can serve a whole small business without one model doing everything.

The AI Profit Boardroom trainings show this per-job model setup inside the Agent OS, step by step.

The Shortlist, With Live Scores

Averages, task counts and medals are from the all-models leaderboard.

Client Pages: The Scores That Matter For An Agency

An agency that has Hermes draft landing pages for clients should look at GoldieBench's Page category, not the games.

Every number comes from that model's own scorecard page, linked in the shortlist above.

For a card such as "draft the florist's spring landing page", Fusion leads the Landing task outright.

Opus 5 and GPT-5.6 Sol both take silver on the promo-page task, so either is a steady pick when the page has to sell an offer.

Pin the pick to that card only, with --model, so the rest of the board keeps its usual model.

The Orchestrator On An Agency Board

The Kanban docs suggest a split: a frontier model on the orchestrator profile that breaks work into cards, and inexpensive models on the worker profiles.

Their reason is that planning needs judgement, while a card with a clear goal and context usually doesn't, and workers use most of the tokens.

From the shortlist, Claude Opus 5 at 8.27 and GPT-5.6 Sol at 8.16 are the single models I'd put on the orchestrator.

Each worker profile then sets its own default in ~/.hermes/profiles/<name>/config.yaml.

Routine Bots: The Salon Inbox And The Shop Report

A salon's inbox Bot and a shop's weekly stock Bot do the same small job over and over.

The Gemini 3.6 Flash page lists "high-volume agentic work" under Best for, and notes builds of roughly 60 to 120 seconds.

Its 7.08 average is the lowest on this shortlist, which is fine for a short summary and not for a client page.

Kimi K3's page points the other way: Best for long-horizon agent runs, with a 1,048,576-token context window, but with 13-plus-minute single builds on the bench.

That suits an accountant's Bot that reads a full year of client files in one sitting, not a quick reply.

For report-only model choices, our Hermes cron job model guide goes deeper.

Per-Channel Models For Team Chat

A shop's Slack can have a quick model in #orders and a stronger one in #suppliers, all from one gateway.

platforms:
  slack:
    channel_overrides:
      "C0ORDERS01":
        model: your-fast-model
      "C0SUPPLY02":
        model: your-frontier-model
        system_prompt: "You handle supplier questions for the shop."

The docs set the order as a session /model first, then the channel override, then the global config.

So a staff member can still switch models for one chat without changing the channel's default.

Using Hermes MoA For High-Stakes Drafts

Hermes MoA scores 8.17 as a panel: one prompt goes to several models, and an aggregator writes the final answer.

Its GoldieBench page lists a default panel of Claude Opus 4.8 and GPT-5.5, aggregated by Opus 4.8, and a latency of roughly 110 to 140 seconds per build.

That fits a one-off proposal for an agency pitch better than a Bot answering staff all day.

Hermes also accepts an MoA preset as a fallback entry, per the fallback providers page, so a panel can step in when the main model is down.

A Model Map For Four Businesses

Each row starts with one job, so the salon or shop can stop after its first line and still have a working setup.

Re-check these picks whenever GoldieBench adds a model, because the leaderboard moves.

Safety Stays The Same Whatever Model You Pick

The approval rules sit in Hermes, not in the model.

The security docs set cron_mode, single_query_mode and unattended_mode to deny by default, so unattended runs on any model have dangerous commands blocked.

The credential vault fills passwords into the page without any model seeing them, and allowlists decide who can message the agent at all.

Pick the model for the work; the guardrails come with Hermes.

Members of the AI Profit Boardroom get these model profiles already set up in the Agent OS.

Also On Our Network

• Agent Operatorsthe ten-step build order, from one cron report to a team of Bots
• AI Income Deskseven owner-friendly use cases, the subscriptions that work and the risks
• AI Tool Verdictten small-business use cases scored out of 10, with the reasons
• agentos.guidefive copy-paste prompts that grow one job into a business setup
• aiprofitboardroom.comthe complete Hermes Agent for business guide on the AI Profit Boardroom blog

Pin one model to your first job today, and book a free strategy session if you want help choosing models for Hermes Agent for business.