Grok Bot Free Trial + Which AI Models Should Plan And Check Your Tasks, With Real GoldieBench Scores

A Grok Bot free trial is a fixed pot of usage credit, so the models you pair with it decide how many useful tasks you get out of that pot.

Grok Bot itself has no model picker: the Grok Bot FAQ says it chooses the backend model for each task.

What you can choose is the model that writes your briefs before they reach the Bot, and the model that checks results afterwards.

Every GoldieBench figure on this page was read from live goldiebench.com pages on 3 November 2026, with the page named beside it.

If you want my team to plan which agent jobs to test first, book a free SEO strategy session.

Why the trial makes model choice matter more

The trial section of the FAQ says the credit drains by agent steps and tokens, not by message count.

A 7-day window sits on top, and used credit is never restored.

So a vague brief that sends the Bot down ten wrong paths costs you real trial credit.

A sharp brief written by a strong planner model, outside Grok Bot, keeps each run short.

That is the whole case for pairing models with a trial.

First, check you still have a trial to spend

Release v0.78.1 on the Grok Bot changelog, dated 27 October, says places that offered a free trial now show Get Grok Bot with a unified plan and See Unified Plans.

The FAQ still carried the Start Trial wording on 3 November.

Open your access screen and see which one you get before you plan any model pairing.

The pairing advice below works either way, since paid plans also draw from a usage pool.

What a GoldieBench score tells you here

The methodology page sets the rules: 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 leaderboard ranks models by their average across the tasks they were scored on.

The home leaderboard lists 50 tasks, 33 models and 1310 live demos.

A high average means a model plans well and gets a working result in one pass.

One-pass planning is exactly the habit that saves trial credit.

The planner shortlist, with scores

Any of these five makes a strong brief writer.

Claude Opus 5 and GPT-5.6 Sol carry the widest coverage on the board, with all 50 tasks scored.

Fusion tops the average, and Hermes MoA is a mixture-of-agents setup that ranks third.

Where Grok sits on the bench

The Grok model page lists a 256,000-token context and 5 gold, 1 silver and 1 bronze medal.

The Grok 4.7 page lists a 500,000-token context and 2 silver and 3 bronze medals.

These numbers do not tell you which model answers a given trial task, because the FAQ says Grok Bot picks per task.

They do show the Grok family can plan and ship in one pass, which is useful context for what the Bot is built on.

A three-step pairing that protects the credit

1. Plan: give one of the top-five models your rough goal and ask for a tight Grok Bot brief with a single outcome and a hard stop.
2. Run: paste that brief into Grok Bot during the trial, then check Settings → Usage & Billing once it finishes.
3. Check: hand the result to a second high scorer and ask it to list errors and gaps before you act on anything.

The planner and checker run outside Grok Bot, so they do not touch the trial credit.

Your trial credit then goes only on the work itself, which is what you are trying to judge.

Rules the planner model should follow

Give the planner a fixed template so every brief comes out credit-friendly.

• One outcome per brief, stated in the first line.
• A list of what the Bot should skip.
• A hard stop after the first useful result.
• An instruction to ask before any long or wide job.

The FAQ itself advises starting small and scoping tasks tightly during the trial, so this template just makes that advice repeatable.

A planner that scores well on one-shot tasks tends to follow a template like this without drifting.

The AI Profit Boardroom has ready-made versions of these brief templates.

Picking a checker model

A checker does a different job from a planner: it reads output and flags problems.

I would pick a checker from a different family than the planner, so the two do not share the same blind spots.

For example, plan with Claude Opus 5 at 8.27 and check with GPT-5.6 Sol at 8.16, or the other way round.

Qwen 3.8 sits at 8.10 across 45 scored tasks on the home leaderboard, which makes it another option.

What no benchmark covers in this trial

GoldieBench scores one-shot HTML builds, not screen control, approvals or billing flows.

So no score helps with the rules that actually catch people out.

• The trial never turns into a paid plan, per the FAQ.
• Cancel Trial removes the remaining credit and cannot be undone.
• An iPhone App Store trial is run by Apple and converts unless cancelled in Apple ID → Subscriptions, per the mobile docs.
• Android has no in-app trial.

The phone side is covered in more detail in the Grok Bot iOS model guide.

How to read Grok's 43 of 47

The home leaderboard shows Grok with 47 tasks on the bench and 43 of them scored.

Its 8.09 average comes from those 43 scored tasks only.

Fusion and Hermes MoA have every one of their 47 tasks scored, so their averages cover the full set.

Keep that in mind when you compare a few hundredths of a point.

After the trial: plan names, not model names

Once the trial ends, your choice is a plan rather than a model.

The Grok Bot on The X Subscription page names Premium, Super and Ultra as the plans that include it; Plus does not.

Super and Ultra give larger shared weekly allowances, which matters if your trial showed you need long agent runs.

Keep your planner and checker models either way, since a tight brief saves usage on a paid pool too.

For the full Agent OS setup around these models, join the AI Profit Boardroom.

Also On Our Network

• Agent Operatorsthe five-step setup for claiming, tracking and cancelling the grok bot trial
• AI Income Deskhow a shop, a service firm and an agency should spend the trial week
• AI Tool Verdictan eight-point scorecard rating the grok bot trial from the docs
• agentos.guidefive agent os prompts that log, scope and judge the trial week
• aiprofitboardroom.comthe complete Grok Bot free trial guide on the AI Profit Boardroom blog

Choose a planner, choose a checker, and book a free SEO strategy session if you want my team to map the jobs for your grok bot free trial.