Grok Bot Plugins + Which AI Models To Pair With Them: Real GoldieBench Scores For Grok, Grok 4.7 And The Top Of The Board

Grok Bot plugins don't let you choose the model behind them, so the real model question is which jobs to leave with Grok Bot and which to send to a model you pick yourself.

The Grok Bot FAQ is blunt: there is no model picker, and Grok Bot chooses the backend model for each task.

The plugin side, meanwhile, just got bigger: since v0.71.2 (14 October 2026), the cap on plugin skills a single Bot can draw from rose from 24 to 200.

Below are the GoldieBench scores I read on 31 October 2026, what they measure, and how I'd split the work.

If you want this mapped to your own stack, book a free SEO strategy session with my team.

Why the model question has a fixed answer inside Grok Bot

Grok Bot picks the model per task and says it can use a first-party SpaceXSI model or a third-party model from an approved provider.

So no GoldieBench score tells you exactly which model ran your Gmail or Linear plugin call.

What the scores can tell you is how the Grok family performs when you call it directly, and which models to run beside Grok Bot.

Plugins themselves are model-agnostic plumbing: OAuth tokens stay on the connector backend, per the computer and apps docs.

What GoldieBench measures, and what it doesn't

The GoldieBench leaderboard gives every model the same prompt, one shot, single HTML file out, scored 0 to 10.

On 31 October 2026 the home page listed 1310 live demos, 50 tasks and 33 models.

That makes it a test of build quality: games, sims, visuals and pages.

It isn't a test of plugin calling, inbox triage or tool selection.

Read every number below as “how well does this model build in one shot”, not “how well does it use Slack”.

Grok on GoldieBench: 8.09 average across 43 scored tasks

The Grok model page lists its best use as workflows that need live X context.

Its own trade-off line says API access is gated behind X Premium, which it calls awkward for backend agent loops.

That's the honest tension: Grok is strong at X-native, fast work, and Grok Bot is where that model family lives most naturally.

Grok 4.7: 7.15 average and rank #18 of 26

The Grok 4.7 model page shows a 7.15/10 average over 20 scored tasks.

It says Grok 4.7 won 14 of 20 head-to-heads against Grok 4.6, which averaged 5.95 on the same briefs.

The Grok 4.7 deep dive, refreshed 2026-10-27, puts it at board rank #18 of 26 ranked frontier models.

Its weakest builds were Voxelcraft at 2.2 and Crypt at 2.3, both scenes that never rendered.

The same model page lists xAI-reported Terminal-Bench 4.0 at 38.0% and DeepSWE v1.1 at 71.0%.

The top of the board, for builds you route elsewhere

All five rows come from the goldiebench.com leaderboard, read on 31 October 2026.

The home page notes Fusion is currently graded by Claude's AI judge, so treat its lead as provisional.

The gap from #1 to #7 is 0.5 points, which is small on a 10-point scale.

Grok's strongest builds on the bench

The Grok scorecard on its model page lists every scored task, best first.

The top of that list is worth knowing before you route anything.

• Twilightvale (game): 9.5, gold
• Game (open brief): 9.0, gold
• Rpg (game): 9.0, gold
• Landing Page: 9.0, gold
• Webos (page): 9.0, gold
• Doom (game): 8.5, silver

Two of the five golds are page builds, which matters for business work.

If a plugin task ends with “now turn this into a landing page”, Grok is a credible builder for that last step.

For games and sims, the leaderboard shows other models with more medals, so route those by category.

What the 200-skill change means for model routing

Before v0.71.2, a Bot could only carry 24 plugin skills, so people split work across several Bots.

Now one Bot can hold up to 200, which pulls more app work into Grok Bot and away from hand-built agent loops.

That shifts the routing question.

Fewer jobs need a model you wire up yourself, and the ones that remain are mostly builds, long specs and research.

Those are exactly the jobs GoldieBench scores, so the leaderboard becomes more useful, not less.

Since v0.75.8, a Bot can also use a plugin's skills right away after installing it mid-task, per the changelog.

How I'd pair models with Grok Bot plugins

• Leave app work with Grok Bot. Inbox, calendar, Notion, Slack and Linear jobs run through plugins, and the model choice is Grok Bot's.
• Send one-shot builds to a top-scoring model. Landing pages, sims and games go to a model you pick inside your Agent OS, such as Claude Opus 5 at 8.27.
• Use Grok directly for live X work. That's the model page's own best-for line.
• Use a panel for visuals. Hermes MoA ranks #3 overall and is marked best at Visuals.

The plugins handle the connected apps; the benchmark picks the builder.

What these scores mean for plugin-heavy work

A high one-shot score shows a model can hold a long spec and wire what it promises.

That's useful when a plugin task ends in a built artifact, like a page or report.

It says nothing about whether a model picks the right tool from 200 plugin skills.

For that, the guardrails matter more than the model: narrow Ask first rules, saved skills and routines tested before you rely on them, per the skills and routines docs.

A five-minute routing checklist

1. Is the job inside a connected app? Use Grok Bot and its plugin.
2. Does it end in a single built file? Route it to a top-five GoldieBench model.
3. Does it need live X context? Call Grok directly.
4. Does it send, post, delete or buy? Put it behind an Ask first rule, whichever model runs it.
5. Check your plan: Grok Bot is on Premium, Super and Ultra, and Plus doesn't include it.

Plan names come from the X Subscription docs.

Want the Agent OS that dispatches these models from one board? Join the AI Profit Boardroom.

Or grab a free strategy session and route your first five jobs with us.

Also On Our Network

• Agent Operatorsthe four-step install order plus every per-app gotcha in the docs
• AI Income Deskthree small-business setups and the risks to plan for first
• AI Tool Verdictan eight-criterion scorecard with a 7.8 overall rating
• agentos.guidefive copy-paste prompts for auditing, installing and approving plugins
• aiprofitboardroom.comthe complete Grok Bot Plugins guide on the AI Profit Boardroom blog

Route your app jobs through the Bot, your builds to a top-scoring model, and book a free strategy session to plan the split for your grok bot plugins.