⭐ Get the Agent OS + join 3,400+ founders inside the AI Profit Boardroom → Join AIPB

GoldieBench Blog · 6 min read

Grok Bot For Content Creation + Which AI Models To Pair With It, With Real GoldieBench Scores For Promos, Loops And Landing Pages

Grok Bot drafts and makes filmed-style clips. For motion promos, designed loops and landing pages, these are the models that scored best on the bench.

Illustrated benchmark leaderboard with gold bars beside a film strip and a draft document on a dark navy background

Grok bot for content creation chooses its own model for every request, so GoldieBench scores cannot rate the Bot itself; they help you pick the models for the content pieces the Bot is not built to make.

Every score here was read on live goldiebench.com pages on 9 November 2026 and is quoted with its page.

The Grok Bot facts come from the official files and results docs and use cases page.

Want help matching models to your content plan? Book a free SEO strategy session with my team.

Why GoldieBench cannot score Grok Bot directly

The Grok Bot FAQ says there is no model picker. The Bot picks a model per task, and the choice can change between requests.

So the Grok row on the leaderboard rates xAI's chat model, not what runs inside Grok Bot.

Where scores do help is the gap around the Bot: the motion promo, the animated background, the landing page for a lead magnet.

Those are builds you run in your own tool, with a model you choose.

The content gap the docs spell out

Grok Bot can create images and short video clips when you ask.

Clips run 1 to 15 seconds, with 6 as the default, at 480p unless you ask for 720p, saved as MP4.

The docs then draw a clear line: clips look filmed and are not made for motion graphics, title cards, slideshows, product or app demos, charts, or anything needing exact text, timing or brand assets.

That line maps neatly onto three GoldieBench tasks: a motion-graphics promo, an animated visual loop and a landing page. Each one is a single HTML file built in one shot.

The leaderboard rows worth knowing

ModelAvg / 10TasksPage read
Fusion8.5947/models/
Claude Opus 58.2750/models/
Hermes MoA8.1747/models/
GPT-5.6 Sol8.1650/models/
Claude Fable 58.1047/models/
Qwen 3.88.1045/models/
Grok8.0947/models/grok
Claude Opus 5.57.5750/models/
Grok 4.77.1520/models/

The /models/ page says every model gets the same fixed prompts, one shot each, scored 0 to 10.

LongCat-2.0 also shows 8.12 there, but over only 4 tasks, so I leave it out of any content pick for now.

Grok 4.7's 7.15 rests on 20 tasks, far fewer than the 47 behind Grok's 8.09, so treat that gap as unsettled.

Motion promos: the task Grok Bot's clips skip

The /tasks/aipbpromo task asks for a cinematic motion-graphics video advert with auto-playing scenes, animated stats and an end call to action.

That is exactly the title-card, exact-text work the Bot's clips avoid.

The page reads 26 models tested, 20 of 26 scored, an average of 7.48/10, and Qwen 3.8 as the winner.

ModelPromo scorePage read
Qwen 3.8winner (gold)/tasks/aipbpromo
GPT-5.6 Sol8.3 (silver)/tasks/aipbpromo
Gemini 3.6 Flash7.8/tasks/aipbpromo
Claude Fable 57.5/tasks/aipbpromo
Fusion7.5/tasks/aipbpromo
GLM-5.27.5/tasks/aipbpromo
Grok7.0/tasks/aipbpromo
Kimi K36.5/tasks/aipbpromo

Grok scores 7.0 here, below the task average, while GPT-5.6 Sol takes silver at 8.3.

For a promo with your stats and logo, pair the Bot's script draft with GPT-5.6 Sol or Qwen 3.8 in your own build.

Animated loops: when a filmed clip is the wrong look

Bot clips can loop: share the same image as both the start and end frame.

But a filmed loop is not a designed one. For a branded animated backdrop behind a post or a page, the closest bench task is /tasks/synthwave, a sunset-grid synthwave loop.

The page reads 26 models tested, 21 of 26 scored, an average of 7.90/10, and GLM-5.2 as the winner.

ModelSynthwave scorePage read
GLM-5.29.0 (winner)/tasks/synthwave
GPT-5.6 Sol8.7 (bronze)/tasks/synthwave
Claude Fable 58.6/tasks/synthwave
Fugu Ultra8.5/tasks/synthwave
Gemini 3.6 Flash8.4/tasks/synthwave
Grok8.0/tasks/synthwave

GLM-5.2 leads at 9.0, and Grok lands at 8.0, just above the 7.90 average.

Landing pages for your lead magnets

Content often ends on a page that collects emails. The /tasks/landing task is a modern marketing landing page in one shot.

That page reads 26 models tested, 21 of 26 scored, an average of 8.11/10, and Fugu Ultra as the winner, with Fusion and GLM-5.2 also at 9.0.

Grok also reads 9.0 on that task page, matching the Landing line on the /models/grok scorecard, so four models share the top score.

For a lead-magnet page, any of the four is a sound pick; choose the one already wired into your build tool.

GPT-5.6 Sol reads 8.7 and Claude Fable 5 reads 7.2 on the same task page.

What the Grok model page says about content

The /models/grok page reads 8.09/10 over 47 tasks and describes Grok as used for real-time content workflows that need current X timeline context.

That fits the Bot's own strength: Grok Bot can search public X posts, mentions, trends and news without your login, as the computer and apps docs explain.

Keep the two apart, though. A score for the Grok model is not a score for Grok Bot, whose model you cannot choose.

Gym example: the Bot drafts, a model builds the promo

  1. The gym owner asks Grok Bot for a short script for a new class launch, drafted from the timetable PDF.
  2. The Bot makes a 6-second 9:16 clip that brings a photo of the studio to life, for stories.
  3. For the launch promo with class times on screen, the owner gives the script to GPT-5.6 Sol, which reads 8.3 on /tasks/aipbpromo.
  4. The owner checks both and posts them by hand.

Tutoring business example: a guide page plus weekly posts

  1. A Monday routine has the Bot draft three study-tip posts and a free revision guide outline.
  2. The tutor asks the Bot for a calm 16:9 image for the guide.
  3. The landing page for the guide is built with GLM-5.2, which reads 9.0 on /tasks/landing.
  4. The tutor reviews the copy and publishes.

Agency example: no model choice needed for drafts

  1. An agency runs the documented Campaign team of three Bots, Sign-ups, Launch drafts and Fact check, on Mondays at 9 AM.
  2. The drafts and fact checks happen inside Grok Bot, where the model is picked for you.
  3. Only the animated background for a client page goes to a chosen model, GLM-5.2, the 9.0 winner on /tasks/synthwave.

How to read these scores for content work

  • They are one-shot creative coding builds, so they show which model builds a promo, a loop or a page well, not which writes the best newsletter.
  • Small task counts, like 4 or 20, carry less weight than 47 or 50.
  • No score here describes Grok Bot itself, and the docs give you no model to pick inside it.

For campaign reporting rather than content, see the Grok Bot for marketing model guide.

Compare your model picks with other builders inside the AI Profit Boardroom.

Also On Our Network

Let the Bot draft, let a chosen model build the promo, and book a free SEO strategy session to plan the topics behind your grok bot for content creation.

FAQ

Which model does Grok Bot use for content?

The docs say Grok Bot picks a model per task and there is no model picker, so no benchmark score describes it directly.

Which model scored best on the motion promo task?

On goldiebench.com/tasks/aipbpromo, read on 9 November 2026, Qwen 3.8 is the winner and GPT-5.6 Sol takes silver at 8.3.

How did Grok score on landing pages?

The goldiebench.com/models/grok scorecard lists Grok at 9.0 on the Landing task, level with the top score on the task page.

Why not use Grok Bot clips for a promo with text?

The xAI docs say Grok Bot clips are not made for title cards, motion graphics or anything needing exact text or brand assets. A promo with on-screen stats is better built in your own tool.

The same stack Julian uses

Run this stack yourself.

Every demo on this bench was built inside the Agent Operating System — one prompt, one shot, single HTML file out. The Agent OS, the prompts, the templates, the weekly walkthroughs and 3,400+ founders shipping with it every day all live inside the AI Profit Boardroom.

3,400+founders
258documented wins
38countries
4coaching calls a week