It feels like every week there’s a new AI model to learn, right? Well, just when I was getting comfortable, Anthropic dropped Claude 4! I just saw this incredible video from an AI professional who really pushed the new models to their absolute limit, and I was blown away by some of the results.
The creator explains that there are two new models: Claude Opus 4 (the top-tier powerhouse) and Claude Sonnet 4 (the new, improved free version). The YouTuber immediately put them to the test in some seriously creative ways.
⚙️ The Gauntlet of Tests
This industry pro didn’t just ask it simple questions. He ran a full-on stress test to see where it shines and where it still stumbles. Here’s the breakdown:
- 🐍 The Impossible Coding Challenge: The creator tasked Opus 4 with coding a Python chess game, but with a custom rule: pawns move like bishops. While it built the game and even revised the code based on a screenshot of file names, the final game logic was broken. It’s a tough challenge that still stumps most models.
- 📄 The “Needle in a Haystack” Test: This was insane. The person who shared it uploaded a 180-page Nvidia annual report and asked for a specific director’s compensation from page 53.
Opus 4 nailed it, pulling the exact number out of a massive wall of text. A huge win for document analysis!
- ✨ Image-to-Code Magic: He took a screenshot of a website banner and asked Opus 4 to turn it into HTML/CSS code. It worked beautifully and even recreated the graphic elements. One strange quirk: the free Sonnet 4 model couldn’t upload the PNG screenshot, which was a feature in the previous version.
- 📊 Data Dashboard Creation: The expert uploaded a screenshot of some old Google Analytics data and asked it to create a simple, shareable dashboard. The result was fantastic: a clean, responsive, and perfectly accurate visual dashboard. A major upgrade from older versions.
- 🧠 Reasoning & Web Search: Using the new “extended thinking” feature, he gave it a classic train riddle. It gave a very plausible answer (Western Pennsylvania) and showed its thought process. It also did a great job using web search to compare itself to GPT-4o and Gemini, even pointing out that its own API cost is significantly higher than its competitors.
💡 My Big Takeaways
I think it’s clear that Claude 4 is a serious contender, especially for specific tasks. The ability to flawlessly analyze huge documents is a game-changer. And that dashboard it created was just beautiful.
However, the video also shows its limits. Complex, multi-step logic in coding is still a hurdle, and the YouTuber hit his usage cap on the Pro plan just by running these tests. If you’re a developer, you’ll definitely want to pay attention to the API pricing, as Opus 4 is on the premium end.
This was an awesome first look. For the full deep-dive and to see all these tests for yourself, make sure to watch the original video from the creator!