GoldieBench Blog · 6 min read
Grok Bot For Recruiting + Which AI Models To Pair With It: Live GoldieBench Pages Scores For Careers Pages And Hiring Briefs
GoldieBench scores the models you pair with a hiring Bot, not the one inside it; here is how to read the overall and Pages columns for careers pages and briefs.

Grok Bot for recruiting chooses its own model for every task, so the benchmark question for a hiring team is which models to put around the Bot, not which one runs inside it.
I read the live GoldieBench pages on 6 November 2026 for this page, and every score below is quoted from the page it sits on.
The Grok Bot facts come from the official Grok Bot FAQ and the use cases page.
Want help choosing a stack for your own hiring and growth? Get your Backlink Gameplan with my team, or try Polar Rank free for 3 days.
No GoldieBench number is a Grok Bot number
Grok Bot has no customer-facing model picker; it manages model selection itself, and xAI doesn't say which model handles a request.
So the Grok row on the board scores xAI's chat model on one-shot builds, not the hiring Bot.
Where the scores help is the work around the Bot: the model in your Agent OS that writes the Talent scout brief, and the model that builds a careers page or a job advert page.
What GoldieBench measures, and what it doesn't
The methodology page says every model gets the same fixed prompt, one shot, and must return a single self-contained HTML file.
Julian scores each result from 0 to 10, and the ranking is the average across scored tasks.
Tasks fall into four groups on the leaderboard: Games (23), Visuals (9), Sims (12) and Pages (3).
Nothing on the bench tests reading CVs, writing outreach or scheduling interviews.
Read the scores as a guide to building pages and following a long brief in one go, and nothing more.
The overall board, read 6 November 2026
| Model | Overall avg / 10 | Tasks | Source |
|---|---|---|---|
| Fusion | 8.59 | 47 | leaderboard |
| Claude Opus 5 | 8.27 | 50 | leaderboard |
| Hermes MoA | 8.17 | 47 | leaderboard |
| GPT-5.6 Sol | 8.16 | 50 | leaderboard |
| Claude Fable 5 | 8.10 | 47 | leaderboard |
| Qwen 3.8 | 8.10 | 45 | leaderboard |
| Grok | 8.09 | 47 (43 scored) | leaderboard |
The top seven sit within half a point of each other.
For writing a hiring brief in your Agent OS, any of them is a reasonable pick; the brief's boundaries matter more than the model.
The Pages tab is the closest match for a careers page
Pages is the smallest category, with three tasks, and each one is a single-file web page.
That is the nearest thing on the bench to a careers page or a one-page job advert.
| Model | Pages avg / 10 | Pages tasks |
|---|---|---|
| Fusion | 8.50 | 3 |
| Claude Opus 5 | 8.47 | 3 |
| GPT-5.6 Sol | 8.47 | 3 |
| Qwen 3.8 | 8.40 | 3 |
| Grok | 8.33 | 3 |
| Claude Opus 5.5 | 7.80 | 3 |
| Claude Fable 5 | 7.57 | 3 |
All seven figures come from the Pages tab on the leaderboard.
Three tasks is a thin sample, so treat a gap of a tenth or two as noise.
Grok's page tasks, one by one
The Grok deep dive lists every scored task.
Its three page builds score Landing 9.0 with a gold medal, Webos 9.0 with a gold medal, and Aipbpromo 7.0.
So Grok tops the field on two of three page briefs and trails on the third, which is why its Pages average lands at 8.33.
If you want a careers page drafted by the model closest to the X ecosystem, that is the record to look at.
Grok 4.7 has no Pages score yet
The Grok 4.7 page shows a 7.15 average over 20 tasks, and all 20 are game builds.
The same page says Grok 4.6 averaged 5.95 on those twenty briefs.
Neither has a Pages result, so don't pick Grok 4.7 for a careers page on the strength of its bench numbers.
A note on conflicting Grok rankings
The live pages don't fully agree about Grok.
The leaderboard ranks it seventh with 43 of 47 tasks scored.
The deep dive's headline box says rank 8 of 26 ranked models.
The methodology page still lists Grok among the unranked models with zero scored tasks.
I've used the leaderboard averages here, since that page holds the per-category data.
If the pages are brought into line later, the ranks may move, but the per-task scores quoted above come straight from the deep dive and the Pages tab.
A model plan for a small hiring team
| Hiring job | Where it runs | What to look at |
|---|---|---|
| Talent scout brief and Ask first rules | Your Agent OS | Overall average |
| Careers page or job advert page | Your Agent OS | Pages tab |
| Sourcing, screening, drafts, slots | Grok Bot | No score applies; Grok Bot picks the model |
Example: a physio clinic hiring a receptionist
- The owner asks her Agent OS, running a top-overall model, for a Talent scout brief that says a person makes every hiring decision.
- She asks a Pages-strong model for a one-page job advert and publishes it herself.
- Grok Bot receives applications at a +jobs alias on its mail.grokbot.com address and logs each one to a sheet.
- Replies wait as drafts until she presses Send, as the email docs require.
- She chooses who to interview, and the Bot proposes Google Calendar slots for her approval.
Example: a two-person web agency sourcing a developer
- The founders run the docs' Talent scout prompt in Grok Bot, which drafts outreach and contacts nobody.
- They approve the messages worth sending; first-contact emails go to one person each with an unsubscribe link.
- For a take-home task, they have a strong overall model write a clear one-page brief.
- Both founders review the work and make the call together.
How to pick a model for your own careers page
Don't stop at the averages.
Open the Pages tab, then open the task pages for the builds closest to your need, and play the demos yourself.
Every demo on the bench is a live one-shot result, so you can judge layout and copy with your own eyes.
Then give your shortlisted model one brief for your careers page and compare it with a second model's attempt.
If neither is right first time, the gap between them on your brief tells you more than a tenth of a point on the board.
Want a second pair of eyes on the result? Get your Backlink Gameplan from my team and we'll review the page and how it shows up in search, or try Polar Rank free for 3 days.
Guardrails no model choice replaces
- Email from a routine waits as a draft, with no option for unattended sending.
- Your own Ask first rules always pause matching actions for a card, per the approvals page.
- The docs tell the Talent scout to respect candidate privacy and each source's terms.
- Plans with Grok Bot: Premium, Super and Ultra, not Plus.
For the sales version of this model plan, read our Grok Bot for sales model guide, and for hands-on help, the AI Profit Boardroom runs the coaching calls.
Pick the models around the Bot from the live board, keep a person in every hiring decision, and join the AI Profit Boardroom to run Grok Bot for recruiting.
FAQ
Which model does Grok Bot use for recruiting tasks?
Grok Bot manages model selection itself, and the docs say there is no customer-facing model picker, so GoldieBench scores tell you about models in general rather than about Grok Bot.
Does GoldieBench test CV screening or outreach writing?
No. It scores one-shot single-file builds in Games, Visuals, Sims and Pages, so use it for page-building and brief-following choices only.
What are Grok's page scores?
On the Grok deep dive read 6 November 2026: Landing 9.0 and Webos 9.0, both gold, and Aipbpromo 7.0, for a Pages average of 8.33 on the leaderboard.
Which model leads the Pages tab?
Fusion, at 8.50 across 3 Pages tasks, with Claude Opus 5 and GPT-5.6 Sol on 8.47, as read on the leaderboard on 6 November 2026.
- 📺 Video notes + links to the tools
- 🆓 Get a FREE AI course + community + 1,000 AI agents
- 🎥 Learn how I make these videos


