GoldieBench Blog · 6 min read
Grok Bot Microsoft Teams + Which AI Models To Pair: What Real GoldieBench Scores Mean For A Shared Bot In Teams
Grok Bot picks its own model, so the GoldieBench question is which models build the brief, guides and trackers around your Teams rollout.
Grok Bot Microsoft Teams answers come from whatever model Grok Bot picks for the task, so the useful benchmark question is which models you run around the Bot, not which one sits inside it.
This page reads the live GoldieBench pages on 11 November 2026 and shows what those scores can and can't tell a team putting a shared Bot into Teams.
The Teams facts come from xAI's shared Bots docs: the channel is for a team's shared Bots, a Microsoft Teams admin must approve each one, every teammate links their own account, and nothing documents it joining meetings or calls.
Want help choosing models for your own rollout? Book a free SEO strategy session with my team.
Grok Bot has no model picker
The Grok Bot FAQ says it chooses the model for each task and offers no picker.
The security page repeats that model selection is managed for you.
xAI doesn't say which model answers a given Teams message.
So no row on GoldieBench is a Grok Bot score, and the Grok row below must not be read as one.
Where the board helps is the work beside the Bot: drafting its brief, and building the pages your staff open to learn how to use it.
What GoldieBench actually measures
The methodology page sets one fixed prompt per task, one shot, with a single self-contained HTML file as the output.
Each result gets Julian's own 0 to 10 score, and a model's rank is its average across the tasks it was scored on.
The tasks are games, visuals, simulations and web pages.
Nothing on the bench tests answering staff questions in a chat thread, so treat it as a guide to build quality, not chat quality.
The overall leaderboard, read 11 November 2026
| Rank | Model | Avg / 10 | Tasks scored |
|---|---|---|---|
| 1 | Fusion | 8.59 | 47 of 47 |
| 2 | Claude Opus 5 | 8.27 | 50 of 50 |
| 3 | Hermes MoA | 8.17 | 47 of 47 |
| 4 | GPT-5.6 Sol | 8.16 | 50 of 50 |
| 5 | Claude Fable 5 | 8.10 | 47 of 47 |
| 6 | Qwen 3.8 | 8.10 | 45 of 45 |
| 7 | Grok | 8.09 | 43 of 47 |
Every row is from the GoldieBench leaderboard, which lists 50 tasks and 33 models that day.
The top seven sit within half a point of each other, so small gaps mean little.
Two more rows sit just outside that group: MiniMax M3 at 7.97 and Fugu Ultra at 7.94.
LongCat-2.0 shows 8.12 in the header ticker, but the leaderboard marks it provisional with four tasks, so it isn't ranked.
The Pages column is the closest fit for a Teams rollout
A rollout needs a staff guide, a link-your-account page and maybe a status tracker, and those are single-file pages.
The Pages category has only three tasks, so read it as a hint.
| Model | Pages avg / 10 (3 tasks) |
|---|---|
| Fusion | 8.50 |
| Claude Opus 5 | 8.47 |
| GPT-5.6 Sol | 8.47 |
| Qwen 3.8 | 8.40 |
| Grok | 8.33 |
| MiniMax M3 | 8.17 |
| Hermes MoA | 8.00 |
The leaderboard's Best at column tags GPT-5.6 Sol with Pages.
Why the Grok row is not a Grok Bot score
The Grok model page scores xAI's Grok model on the same one-shot builds as everyone else.
Its Pages builds score Landing 9.0, Webos 9.0 and Aipbpromo 7.0.
That says how Grok handled those prompts on the bench, not how Grok Bot replies in a Teams channel, because Grok Bot manages its own model choice.
Conflicts on the bench you should know about
The methodology page still lists Grok among the unranked models with zero scored tasks.
The Grok model page has a trade-off line saying its demos have no curated verdicts yet.
Yet the leaderboard ranks Grok seventh at 8.09 with 43 of 47 tasks scored.
The counts differ too: the home page shows 1041 curated verdicts, while the methodology page says 1259 of 1609 cells.
Go by the live leaderboard row and its date, and recheck before you quote it.
A pairing for an IT support firm's Teams rollout
- In your Agent OS, a top-ranked model drafts the shared Bot's brief: purpose, limits, "draft only".
- A strong Pages model builds a one-page staff guide: what to ask, how to link your Teams account under Settings, General, Shared Bots.
- The owner enables Teams on the dashboard, connects the Bot, and the Teams admin approves it.
- The shared Bot then answers technicians in the Support channel.
A pairing for a small agency's account team
- A model in your Agent OS builds a single-file tracker listing who has linked, copied from the dashboard's Manage Microsoft Teams accounts list.
- Another drafts the pinned welcome post, saying clearly the Bot doesn't join meetings or calls.
- Account managers @mention the Bot for briefs; it replies in the thread.
For the wider shared-Bot setup, see the Grok Bot teams breakdown.
A pairing for an accountancy practice
- A model in your Agent OS drafts the Bot's limits: office-process answers only, draft only, no client contact.
- A Pages model builds a single-file month-end checklist page that staff can open beside Teams.
- The shared Bot answers process questions in the Office channel and points staff to that page.
- Client records stay out of the Bot's files, because channel replies are visible to the channel.
How to test a model for your rollout pages yourself
- Write one fixed prompt for your staff guide, the way the methodology fixes one prompt per task.
- Run it once on two or three models from the Pages table.
- Open each HTML file in a browser and check it on a phone, since staff will open it from Teams.
- Keep the one your staff understand fastest, and save the prompt for the next update.
That mirrors how the bench works, but on your own content, which no leaderboard can do for you.
What no benchmark tells you about Teams
- Whether your Teams admin will approve the app, or whether a policy blocks custom apps.
- How approval cards look in Teams: the docs explain them for Slack and group chats, not Teams.
- Which computer the Bot uses there. All of a user's Bots share one cloud computer, with each Bot on its own screen, but the docs give no Teams-specific detail.
- File handling inside Teams, which isn't documented.
Add Ask first rules from the approvals guide to cover those gaps.
The bottom line
Use the leaderboard to choose the models that build your rollout pages and draft the brief.
Leave the Teams answers to the shared Bot, and judge those on real staff questions.
Members of the AI Profit Boardroom post their model picks for jobs like this every week.
Open the leaderboard, pick one Pages model and build your staff guide today.
Also On Our Network
- Agent Operatorsthe five-step Teams setup with a fix for every stuck status
- AI Income Deskwhether a shared Bot in Teams is worth it for a small team
- AI Tool Verdicta 6.6/10 scorecard for shared Bots inside Microsoft Teams
- agentos.guidefive Agent OS prompts that carry a Teams rollout's paperwork
Want a plan for models and Bots together? Book a free SEO strategy session or join the AI Profit Boardroom, and pair the right models with Grok Bot Microsoft Teams.
FAQ
Which model does Grok Bot use in Microsoft Teams?
xAI doesn't say. Grok Bot picks the model for each task and has no model picker.
Is the Grok row on GoldieBench a Grok Bot score?
No. It scores xAI's Grok model on one-shot builds, not Grok Bot's replies in Teams.
Which GoldieBench category matters most for a Teams rollout?
Pages, because staff guides and trackers are single-file pages, though it only has three tasks.
Does the benchmark say anything about Teams meetings?
No, and neither do the Grok Bot docs. The Teams channel covers chats and channel mentions only.
Keep reading
Grok Bot Teams + Which AI Models Matter: What Real GoldieBench Scores Tell A Team Using Shared Bots
Grok Bot Phone Calls + Which AI Models To Pair With Them, With Real GoldieBench Scores
Grok Bot Connectors + Which AI Models Should Handle Your Connected App Data, With Real GoldieBench Scores🌐 Sister-site take: read this on agentos.guide