Hermes Agent For Real Estate + Which AI Models To Run For Property Pages, Listing Drafts And Private Client Notes, With Real GoldieBench Scores

Hermes Agent for real estate lets each job on a property desk run on a different AI model, so the useful question is which model belongs on the development microsite, which on the listing drafts and which on the private client notes.

Every score below is one I read on the live GoldieBench pages on 8 November 2026, with the page linked beside it.

The examples are a lettings office, a new-homes sales team and a solo buyer's agent, and every client message stays draft only, since Hermes has no approval step for gateway replies.

Want my team to pick models for your own agency workflow? Book a free SEO strategy session.

What GoldieBench Measures, And What That Means For An Agency

The GoldieBench methodology gives every model one fixed prompt, one attempt, and the output has to be a single self-contained HTML file.

Julian scores each result from 0 to 10, and the leaderboard ranks models by their average over the tasks they were scored on.

For a property business, one task maps closely: Landing, a one-shot modern marketing landing page.

A one-page site for a new development or a single flagship home is the same class of build.

GoldieBench does not score listing descriptions, enquiry replies or CRM lookups, so for those jobs the overall averages are a rough signal of general strength and nothing more.

Landing Scores: The Closest Match To A Property Page

The Landing page shows 26 models attempted, 21 scored, an average of 8.11 out of 10, and Fugu Ultra named as the winner.

Four models share the top score of 9.0, so any of them is a fair first pick for a development page.

The Fable 5 note on the task page flags a 3D object overlapping card text, which is exactly the kind of flaw a buyer would notice.

Overall Averages For The Writing-Heavy Jobs

Listing copy and enquiry drafts aren't benched, so here are the leaderboard averages from the models page as a general guide.

These are build scores, not copywriting scores, so treat a gap of a tenth or two as noise for writing work.

Five Places Hermes Lets You Set The Model

• The main model, chosen with hermes model, used by anything without its own pin.
• Each scheduled job, pinned with --model and --provider, or locked to the current model with --pin, per the cron docs.
• Each Bot, through the model and provider pin in Bot Mode, so a Property Desk Bot can differ from your sales Bot.
• Each chat channel, with channel_overrides in the messaging gateway config.
• A single prompt, sent through the default Mixture of Agents preset with /moa, per the slash command reference.

Matching Models To Five Property Desk Jobs

The picks are my reading of the scores, not a GoldieBench ranking for these exact jobs.

Start with the job you run most often, pin a model to it, and compare a week of drafts before you change anything else.

If the edits your team makes keep shrinking, keep the pin; if they don't, try the next model in the table.

Hermes MoA: When One Prompt Deserves A Panel

The Hermes MoA page lists an 8.17 average across 47 scored tasks.

It describes a default panel of Claude Opus 4.8 and GPT-5.5, merged by Opus 4.8, run from the Mixture tab in Julian's Agent OS.

The same page lists the trade-offs: roughly 110 to 140 seconds per single-file build, and more calls per task than any single model.

In Hermes itself, the Mixture of Agents preset runs one prompt through a panel and then returns you to your session model.

Save it for the one page or letter that matters most, such as the launch page for a new scheme.

Local Models For Notes That Must Stay In The Office

The local models board covers seven strictly on-device models run through Ollama or MLX, where nothing leaves the machine.

Its leader is Qwable 5 27B Coder at 7.14, followed by Agents-A1 at 4.83 and Gemma-4 12B Coder at 4.25.

That board ranks coding builds, and the gap to the cloud averages above is wide.

Hermes can point at self-hosted endpoints such as Ollama and vLLM, per the providers page.

So a sensible split is a local model for sorting private notes and a cloud model for public-facing copy that a person checks anyway.

Three Agencies, Three Model Setups

A lettings office pins its weekday chaser job to Gemini 3.6 Flash.

1. The job reads quiet leads from the CRM each morning.
2. It writes a two-line draft for each one, marked draft only.
3. The list lands in the team chat and staff send what they approve.

A new-homes sales team gives its Property Desk Bot Claude Opus 5 for descriptions.

1. The manager emails the spec PDF and photos to the Bot.
2. The Bot drafts a full description and a portal summary with every detail flagged for checking.
3. For the launch page, she runs one /moa prompt and compares it with the GLM-5.2 version.

A solo buyer's agent keeps client notes on a local model.

1. Her Bot on a self-hosted endpoint sorts private notes and viewing feedback.
2. Area briefs from public pages run on a cloud model through a separate scheduled job.
3. She checks the links and sends each brief herself.

Members of the AI Profit Boardroom share their own model pins for setups like these.

What No Benchmark Fixes

A stronger model doesn't add an MLS, Zillow or portal integration; Hermes has none in either catalog.

It doesn't make a listing claim true; a person still checks every measurement and feature.

And it doesn't add an approval step for gateway replies, so draft only stays in every job.

For the outbound side, see our Hermes Agent for sales model guide.

Want help tuning the stack? The AI Profit Boardroom covers model picks on its weekly coaching calls.

Also On Our Network

• Agent Operatorsthe seven-step lead desk build, from profile to weekday call list
• AI Income Deskfive small-agency use cases and the risks an owner should weigh
• AI Tool Verdictthe ten-row scorecard, from channels at 8 to property data at 2
• agentos.guidefive copy-paste prompts that set up a Property Desk Bot
• aiprofitboardroom.comthe complete Hermes Agent for Real Estate guide on the AI Profit Boardroom blog

Pin one model to one job this week, check the drafts, and book a free strategy session if you want help choosing models for Hermes Agent for real estate.