Hermes Webapp + Which AI Models To Run In Your Browser Sessions, With Real GoldieBench Scores For Opus 5, GPT-5.6 Sol, Kimi K3 And Grok

The Hermes webapp changes where you reach your agent, not which model does the thinking, so the model you pick still decides how good every browser session is.

The Webapp shipped in Hermes Agent v0.22.6 on 2 November 2026, and it serves the full Desktop workspace to a browser from the gateway.

Below are the GoldieBench scores I read on the live models page on 3 November 2026, which models suit which kind of Webapp work, and how profiles let you switch between them from the tab.

If you'd rather my team matched models to your workload, book a free SEO strategy session.

The Model Runs On The Gateway, Not In The Tab

The Hermes Desktop docs say chats in the Webapp run on the gateway, the same process that serves the CLI, the native Desktop and messaging.

So the browser never runs a model of its own.

Whatever model your profile is set to in the gateway is the model every Webapp chat uses.

That keeps the choice in one place: change it once and every surface follows.

It also means a slow model feels slow everywhere, in the tab, the CLI and your bots alike.

The upside is that the score you read below applies to every one of those surfaces at once.

The GoldieBench Shortlist For Webapp Sessions

The models page compares 33 frontier models, and Fusion tops it at 8.59 across 47 tasks.

Fusion is listed as an OpenRouter multi-model panel, so I've left it off a table of single models you'd set on a Hermes profile.

What A GoldieBench Score Tells You Here

GoldieBench gives every model the same fixed prompts, one shot each, scored 0 to 10 by Julian.

Each task asks for a single HTML file, which makes the score a read on first-try build quality.

That maps well to Webapp work, because the workspace puts previews, the file tree and Git review side by side with the chat.

A model that builds it right first time saves you a round of review in that Git pane.

It doesn't measure how a model behaves with host access, and that's a point worth taking seriously in a tab that can run commands on your machine.

Heavy Builds From The Browser: Opus 5 Or GPT-5.6 Sol

For big jobs you start from the Webapp and review in its preview and Git panes, the top two single models are close.

Claude Opus 5 averages 8.27 over 50 tasks, and its page lists a 1,000,000-token context window.

GPT-5.6 Sol averages 8.16 over 50 tasks, with a context window its page gives as 1,050,000 tokens.

The 0.11 gap is small, so pick the one your provider setup already supports.

Long Unattended Runs: Kimi K3

Kimi K3 scores 7.89 across 50 tasks, and its page describes it as tuned for long-horizon agent work, with a 1,048,576-token context.

That suits a job you start from the browser, leave running, and check later from another device.

The terminal rail fits that pattern too: the docs say detached shells are kept for 15 minutes while the gateway runs, and reconnecting replays up to 1 MiB of output.

Fast Everyday Chats: Grok Or Gemini 3.6 Flash

Grok averages 8.09 over 47 tasks, and its card on the models page calls it snappy and real-time.

Gemini 3.6 Flash scores 7.08 over 50 tasks, and its card sums it up as faster, cheaper and using fewer tokens.

For quick checks from a phone between meetings, a fast model keeps the tab responsive.

Panel Answers: Hermes MoA

Hermes MoA averages 8.17 over 47 tasks.

Its page describes a panel of frontier models whose drafts are merged by an aggregator, with a default panel of Claude Opus 4.8 and GPT-5.5.

It's a system rather than a single model, so treat it as the option for answers where a second opinion is worth the extra calls.

Switch Models With Profiles From The Tab

The CLI reference describes profiles as isolated Hermes instances, each with its own config.

That makes a profile per model lane the clean setup: one for heavy builds, one for long runs, one for fast chats.

hermes -p work webapp opens the Webapp on the work profile, and profiles the default gateway serves share its Webapp.

The v0.22.6 notes also list /model reset, which clears a session's model pin without losing history.

Use it when a session you pinned to a heavy model should drop back to the profile's default.

A Three-Profile Setup By Score

This is an editorial pairing from the scores and the model cards, not a GoldieBench ranking of agent work.

Open each with hermes -p build webapp, hermes -p long webapp or hermes -p quick webapp.

Display preferences sit in one ui-preferences.json on the host, so every profile's tab keeps the same look.

Members of the AI Profit Boardroom get the installable Agent OS with Hermes profiles already set up, which makes this split a quick change rather than a project.

Turn It On Before You Compare

These are the commands from the docs.

hermes config set webapp.enabled true
hermes gateway restart
hermes webapp

It opens on http://127.0.0.1:9120 by default, using a one-use sign-in link valid for 60 seconds.

The Webapp is not the Web Dashboard, which runs separately with hermes dashboard on port 9119.

To open it from another device, use Settings → Web Access in the native app, behind your Nous account sign-in.

Prefer the HTTPS My own URL option, saved as webapp.public_url, since plain HTTP away from the host turns off the microphone, clipboard reading, notifications and the bot screen viewer.

Never forward the port from your router, and remember that a signed-in user can run commands as the account running Hermes, whichever model is loaded.

Plugins Don't Change The Model Pick

The Webapp loads desktop plugins from the host's desktop-plugins folder, so the tab gets the same extensions as the native app.

Plugins add tools and panels around the model rather than replacing it, so a strong model with the right plugins beats a weak one with many.

For which plugins to add, see the best Hermes Agent plugins on this site.

Want the full Agent OS around these models? Join the AI Profit Boardroom.

Also On Our Network

• Agent Operatorsthe six-step setup for the Hermes Webapp, from config to remote sign-in
• AI Income Deskwhether a small business should run its Hermes agent from a browser
• AI Tool Verdicta nine-point scorecard rating the Hermes Webapp, strengths and limits
• agentos.guidefive copy-paste prompts to run your Agent OS through the Hermes Webapp
• aiprofitboardroom.comthe complete Hermes Webapp guide on the AI Profit Boardroom blog

Pick a lane from the table, give it its own profile, and if you want my team to check the pairing, book a free strategy session before you roll out the Hermes webapp.