For the last two years, I’ve been comparing AI models like some people compare sports teams. I’ve used ChatGPT since GPT-3. I’ve watched Google stumble with Bard, then Gemini, then Gemini 2.0. I’ve seen startups like Anthropic and DeepSeek eat away at the edges. But the main event has always been OpenAI vs Google.
This week, the gloves came off. Google released Gemini 3.0 on June 18, claiming it’s “the most capable AI in the world.” Two days later, OpenAI pushed out GPT-5.1 with new multimodal features. I spent the last seven days running them through a gauntlet of tests—writing, coding, reasoning, and real-world tasks. Here’s what I found.
The Setup: Same Hardware, Same Prompts
I used the web interfaces for both: ChatGPT Plus ($20/month) and Gemini Advanced ($22.99/month as part of Google One AI Premium). Both offer the latest models. I ran each test three times and took the best response. The tests were designed to reflect how normal people—not AI researchers—actually use these tools.
Test 1: Writing a Blog Post
Prompt: “Write a 500-word blog post about why houseplants are good for mental health. Use a conversational tone.”
ChatGPT’s response was solid. It opened with a personal anecdote (“I bought my first snake plant during a rough patch...”), hit the science lightly, and ended with a call to action. It was well-structured and easy to read. But it felt like a template—I could almost see the formula behind it.
Gemini 3.0’s response surprised me. It was more specific. It mentioned a 2025 study from the University of Exeter about cortisol reduction. It included practical tips like which plants are best for beginners (pothos, snake plant) and which to avoid if you have pets. It felt like someone who actually owns houseplants wrote it. The tone was warm but not cloying.
Winner: Gemini. ChatGPT’s writing is reliable, but Gemini’s felt more genuine.
Test 2: Coding a Simple App
Prompt: “Write a Python script that uses the YouTube API to download the top 10 trending videos’ titles and view counts, then saves them to a CSV.”
ChatGPT produced a working script on the first try. It included error handling, comments, and a function to handle API rate limits. It was clean and professional.
Gemini’s script also worked, but it used an outdated API endpoint. I had to fix two lines. However, Gemini’s explanation of how the API works was much better—it explained OAuth authentication and pagination in a way that a beginner could understand.
Winner: ChatGPT for code quality. Gemini for explanation.
Test 3: Analyzing a PDF
I uploaded a 50-page PDF of a research paper on microplastics and asked both to summarize the findings and identify the three most important statistics.
ChatGPT handled this beautifully. It parsed the document accurately, extracted key numbers (e.g., “microplastics found in 93% of bottled water samples”), and presented them in bullet points. No errors.
Gemini 3.0 has a new feature: it can highlight specific sections of the PDF and cite them inline. So when it said “microplastics found in 93% of samples,” it showed me the exact page and paragraph. That’s a game-changer for research. But it missed one statistic that ChatGPT caught.