It's been three and a half years since ChatGPT launched and changed everything. In that time, we've seen an explosion of AI assistants — Google's Gemini, Anthropic's Claude, Meta's Llama, plus a dozen smaller players. Each claims to be the best. Each has its passionate defenders. But which one should you actually use?
I decided to find out. I spent the first week of June 2026 testing the four major consumer-available models — ChatGPT (GPT-5), Claude 4, Gemini Ultra 2, and Llama 4 (via a hosted API). I gave each the same tasks: writing, coding, research, creative brainstorming, and everyday productivity. I scored them on accuracy, speed, tone, and creativity. Here's what I found.
The Contenders (as of June 2026)
A quick note on versions. OpenAI released GPT-5 in March 2026, and it's a significant leap over GPT-4. Anthropic's Claude 4 launched in late 2025 and has been steadily updated. Google's Gemini Ultra 2 came out in April 2026 and finally feels like a serious competitor. Meta's Llama 4 is open-source but requires some technical setup to run locally; I used the hosted version via Together.ai for consistency.
All models were tested using their default settings (no custom instructions, no temperature adjustments). I used the paid tiers where applicable: ChatGPT Plus ($25/month), Claude Pro ($25/month), Gemini Advanced ($22/month via Google One), Llama 4 (free via Together.ai but with usage limits).
Task 1: Writing a Persuasive Email
Prompt: "Write a polite but firm email to my landlord about a broken washing machine that hasn't been fixed in two weeks."
ChatGPT: Solid, workmanlike. It hit the right tone — polite, factual, with a clear request for action. But it was a bit boring. Like a template from a business writing course. 7/10.
Claude: Better. It added a subtle emotional touch — "I understand you're busy, but this is starting to affect my daily routine" — without being whiny. It also suggested attaching photos and previous correspondence. 9/10.
Gemini: Oddly formal. It used phrases like "I hereby request" and "pursuant to our previous communication." Too stiff for an email to a landlord. 5/10.
Llama: Surprisingly good. It wrote a short, direct email that felt human. Didn't overexplain. But it had a small factual error — it said "two weeks" even when the prompt said two weeks. Okay for a draft. 7/10.
Winner: Claude.
Task 2: Writing Python Code for Data Analysis
Prompt: "Write a Python script that reads a CSV file, cleans missing values, and generates a bar chart of sales by month."
ChatGPT: Generated a clean, well-commented script using pandas and matplotlib. It included error handling and a check for missing files. It worked on the first try. 9/10.
Claude: Similar quality, but it added a docstring and type hints, which is nice for production code. It also suggested using seaborn for better-looking charts. 9/10.
Gemini: The code was functional but verbose. It used unnecessary try-except blocks and commented every single line. Readable but cluttered. It also imported libraries I didn't need. 6/10.
Llama: The code was concise and correct, but it didn't include any error handling. When I ran it with a deliberately corrupted CSV, it crashed. It's the fastest, but less robust. 6/10.
Winner: Tie between ChatGPT and Claude.
Task 3: Research and Summarize a Complex Topic
Prompt: "Explain the current state of quantum computing research, including recent breakthroughs in error correction, in a way a college student could understand."
ChatGPT: Gave a clear, structured overview with sections on error correction, topological qubits, and recent Google and IBM results (as of early 2026). It cited specific papers and dates. Impressive depth. 9/10.
Claude: Similar depth but written in a more engaging style. It started with a metaphor ("Quantum computers are like trying to solve a maze by walking all paths at once") that was actually helpful. It also flagged areas of uncertainty. 10/10.
Gemini: This is where Gemini shines. It pulled in real-time data from Google Scholar and displayed citations. Its summary was accurate and well-sourced. But the writing was dry. 8/10.
Llama: The summary was too high-level. It got the basics right but missed recent advances in error correction (like the surface code improvements from 2025). Fine for a beginner, but not current. 5/10.