I put Grok vs MetaAI in a 7-round face-off — here's the winner

1. Image Generation

First up we’re going to run the AI image generation test. Grok uses its own built-in model previously known as Aurora. Gemini is using the Google DeepMind Imagen 3 model. Neither are using native image generation, but then no public model uses that technique yet.

Here I’m asking the AI model to generate an image of a man with glasses and a beard at a messy desk. This is not autobiographical, honestly.

The prompt: "A man in his early 40s with a beard and glasses is sitting at a messy desk in a home office, looking intently at a laptop screen with a focused expression. He's wearing a comfortable sweater and jeans. One hand is on the mouse, the other is holding a pen that he's tapping against his chin. There are stacks of books and papers surrounding him, and a half-eaten bowl of ramen sits on a coaster beside the keyboard. Light from the setting sun streams through the window behind him, casting long shadows across the room."

2. Social Media

In all of my previous tests I've made prompt 2 about image analysis but I'm in the UK and MetaAI doesn't have image analysis capabilities. I can't give it a picture, so instead I've created a prompt playing on the social media connection.

Prompt: "Imagine you are a social media manager for a new eco-friendly clothing brand. Write three different social media posts (for platforms like Instagram, Facebook, or X) to announce the launch of the brand and its first collection.

The posts should be engaging, informative, and target a young, environmentally conscious audience. Include relevant hashtags and consider the unique style of each platform."

3. Coding Challenge

In previous tests I’ve had the models create games, a to-do list app and a pomodoro timer. Here I’m getting them to create a simple convertor. In this one we’re converting length and weight.

Prompt: "Create a Python program with a GUI that functions as a simple unit converter. It should allow users to:

Choose between converting length (meters to feet/feet to meters) or weight (kilograms to pounds/pounds to kilograms).

Input a value to be converted.

Display the converted result.

The interface should be clean and user-friendly. The code must be executable without any modifications."

4. Creative Writing

For challenge four we’re going to get each model to generate a short story in the style of Dr Seuss but they have to be about a young inventor creating an animal language translator.

Prompt: "Write a whimsical short story in the style of Dr. Seuss about a young inventor who creates a machine that can translate animal languages. They use it to communicate with their pet goldfish, who reveal a surprising secret about the origins of their species. The story should be filled with rhymes and imaginative creatures."

Full report in a Google Doc. MetaAI's response failed almost immediately for me as it used the word whimsy in the opening line. This is a little "on the nose" and a sign of a lower-quality AI model.

5. Problem Solving

Problem-solving is something AI models can be good at, especially if they have a degree of reasoning capability. They work through the problem step-by-step and provide a solution. The challenge here is in how well they present that solution for a non-technical audience.

Prompt: "A user is having trouble connecting their wireless headphones to their laptop. They have tried turning the headphones on and off, but the problem persists. Develop a troubleshooting guide that covers common connectivity issues, including Bluetooth settings, driver updates, and potential hardware problems."

Full response available in a Google Doc. MetaAI didn't do a bad job. It broke it down step-by-step with simple instructions, even if every one seemed to be "reach out to the manufacturer." Grok just did it better with a more concrete plan and fallback options.

6. Advanced Planning

AI models are very good at planning, especially ones like Gemini and ChatGPT Search that have live data access. As both MetaAI and Grok also have live access I thought I’d see how well they handled planning a vacation to the Scottish Highlands.

Prompt: "Plan a 10-day trip exploring the Scottish Highlands and Islands for a solo traveler interested in hiking, wildlife spotting, and experiencing local culture. The plan should include:

A suggested itinerary with a mix of mainland and island destinations (must include Isle of Skye and Loch Ness).
Recommendations for scenic hiking trails with varying difficulty levels.
Suggestions for opportunities to observe local wildlife (red deer, seals, birds).
A variety of accommodation options (hostels, B&Bs, and unique stays like glamping or bothies).
Transportation suggestions (public transport, car rental, ferries).
Estimated budget breakdown in GBP, considering accommodation, transportation, activities, and meals."

Full details in a Google Doc. Similar responses, but as with the other responses, Grok was more personal and engaging. It was generally better overall with more nuance and detail.

7. Education

Finally, we’re testing how well AI manages to explain a complex topic to a specific audience. Here I’ve asked it to explain the concept of artificial intelligence to a 12-year-old. This has to include a break down by topic and show everyday examples.

Prompt: "Explain the concept of artificial intelligence in a way that a 12-year-old could understand. Use analogies and examples from everyday life. Discuss the different types of AI and how they are being used today."

Full responses in a Google Doc. Grok does a much better job of breaking down the concept, splitting it up by different types of AI and using more engaging analogies.