# Grok Bot vs Claude: Grok Bot Already Hands Some Jobs to Opus 5.5

Channel: Simply AI
Video: https://www.youtube.com/watch?v=06lDAVA9uxA
Duration: 9 min
Language: English
Words: 1535
Transcript page: https://viewrankai.com/tools/youtube-transcript/06lDAVA9uxA

---

[0:00] Pay $30 a month for Grokbot and some of your jobs get done by a rival's model, Claude Opus 5.5 from Anthropic. Grokbot's own help page says so. The same page asks, "Can you see which jobs Claude did?" Its answer is one word, "No." You can't pick Claude or block it, either. If your bot reads your inbox, your email might go to XAI's model or to Anthropic's, and the app never tells you which. There is an upside. When Opus gets the job, the help page says it costs you no more per token. It still comes out of the same weekly allowance, though, and nobody publishes how much of your work Opus

[0:39] gets. That answers the price question, too. Against Claude's own $20 plan, nobody can tell you whether the $30 of Grokbot buys you more Opus. Only paying Anthropic tells you it's Claude you're getting. Later, you'll see what Grokbot does that Claude's own assistant can't. You can watch it work and take over. If you haven't met it, Grokbot is an assistant with its own computer in the cloud, doing the jobs you message it, even with your laptop shut. In early October, Elon Musk said it would use the best back-end model for any given task, and the first one he named was Claude Opus 5.5. You'd expect your bill to at

[1:17] least say which model did each job at that model's price. One early write-up said it would. It doesn't. Your usage screen shows one line, Grokbot usage, and a cheap model and Opus look exactly the same on it. The one choosing is Cursor, the coding app on the same X subscription as Grok and Grokbot. Cursor runs the computers your bots work on, and it picks the models, too. Its whole rule, in its own words, is the model

[1:44] most likely to deliver the best outcome. Which jobs that means, it doesn't publish. There's one more thing about that choice. Nobody can switch it off, not even your company. Companies that use Cursor can keep an allow list, the only AI models their staff are allowed to use. Grok bot doesn't follow it. Its help page says it may use outside models regardless of those settings. Teams get a notice when they switch it on and that's it. If your company kept Claude away from its data, a Grok bot can still send that data to Claude. It was doing that before anyone announced it. In late August, a user on Cursor's forum asked

[2:22] why their credits kept draining. Their usage showed requests to a Claude Opus model they never picked and the draining stopped the moment they paused the bot. Cursor support answered that these are models Grok bot picks on its own servers. That was 5 weeks before Musk's post. People were paying for Claude before anyone told them it was in there and I find that hard to shrug off. Your usage screen no longer shows those names. The list of outside companies can change, too. The security page says

[2:50] there's no fixed vendor set guaranteed. If a model isn't available where you live, the job stops, says why, and falls back to Grok. Each request goes to XAI's model or to one of those providers. Cursor's agreements say the providers don't keep your data with one exception in the fine print. Providers may run safety checks and flagged data may be stored for investigation. With your team's privacy mode on, the docs say your data isn't used for training. No more per token doesn't mean free, either. Grok bot draws from one weekly pool it shares with Grok and Cursor. How fast it empties depends on how much effort each job takes and after that,

[3:31] you buy top-ups. One heavy user on Hacker News says that since starting with Grok bot, their AI spending has gone up two to three times. Fine print like this is most of what we read on this channel. If you'd rather know what's inside the tools before you pay for them, subscribe. That's what you hand over. Here's what people get back for it. An independent reviewer logs real GrokBot runs and posts the evidence. In one, it fixed some broken code, saved the fix, and put a test listing live. A pass. In another, it spent 42 minutes trying to get into X, was refused every time, then stopped and

[4:07] said so. A fail, but an honest one. That Hacker News user is blunt about speed. A job through a GrokBot often takes an hour against about 10 minutes with an AI on their own computer. They still say people would be surprised how helpful it is. On Reddit, one user's grocery bot plans the week's dinners and fills a Kroger cart. They approve it and book the pickup. A job like that runs on a routine, on a schedule, or when something happens, like an email arriving or a Slack message. A schedule can be as tight as every 5 minutes. You can run several bots, each with a name

[4:43] and a job, and put up to six in one group chat where they message each other. In xAI's own demo, a bot called website maker builds a page, then hands it to one called copywriter. My favorite part is that you can just show it. Do the job once while it records you, and the bot can repeat it. On X, you can tag your bot under a post, and it picks the job up. It can even have its own email address. But a routine can't send mail by itself. It leaves a draft and waits for you to press send. When a site wants a password or a CAPTCHA, you open the

[5:16] bot's computer, do that one step, and hand it back. Claude's own assistant won't let you do that, and it's the difference that matters most when the two go side by side. Some people go further and bring Claude in on purpose. The security page says bots can act on your own machine through the desktop app, run commands, and read files. That means a bot can type into Claude code on your Mac, on your own Claude

[5:39] subscription. One Reddit user's verdict? Works great. Making it reliable is harder. One project's own test stopped because its Claude login wasn't working, and another warns that jobs fail until you set up a login token. That route means paying both companies, which brings us back to the price fight. Claude Pro is $20 a month, and it already includes Claude Code, the coding

[6:03] tool those Reddit users were driving. Since October 21st, it also includes Claude Duet, Anthropic's always-on assistant as a preview, US only with a waitlist. Like Grokbot, Duet works on its own computer in the cloud with apps you connect. They split on two things. You can't watch Duet in its browser or take over, and Grokbot lets you do both. With Duet, though, you know which

[6:27] company's model is reading your inbox. Grokbot won't even put its weekly allowance in jobs or hours. ChatGPT's agent is gone. OpenAI's help page says it's no longer available and points people to ChatGPT work for longer jobs from the plus plan. Its always-on assistant is called a dot, and it needs a pro plan from $100 a month. It isn't offered in the EU, the UK, or Switzerland. For that money, your dot starts out knowing what ChatGPT already remembers about you. It can hand coding jobs to Codex, and you can take over its browser, too. You also know what's driving it, OpenAI's own GPT-6 Astra. What only Grokbot gives you is a

[7:09] crew, several named bots instead of one, jobs you tag straight from X, and routines as often as every 5 minutes. ChatGPT's scheduled tasks run at most once an hour. It comes down to what you'd rather have, more helpers or knowing whose model is helping. Grokbot makes sense if you want several helpers working all day from X, Slack, and email. You like watching them and stepping in, and you don't mind not knowing whose model did the work. At $30, it costs far less than a dot, and unlike Duet, there's no waitlist. Skip it if your company keeps a model allow list or your work data can't go to an

[7:47] outside provider. Grokbot follows neither. And if it's Claude you're after, you already know where to go. Don't look for a free way in, either. None of the three is on a free plan. Next, I'm watching the rollout to companies, which Musk reportedly told investors is a few weeks away. Every company that tries it will ask the

[8:05] question this help page answers with no. If XAI ever lets you pick the model, Grokbot gets a lot easier to say yes to. Grokbot's mystery mix for 30 or Claude by name for 20? Which one would you pay for? If this helped, a subscribe really means a lot. I break down the AI news that matters, just like this. And to learn

[8:27] more, watch one of our deep dives next. [music]

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt