Claude Sonnet 4.5 is Here: New Pacman Challenge
Testing Anthropic's latest Claude Sonnet 4.5 model with the same Pacman challenge. Unlike Opus 4.1 and GPT-5, it didn't nail it on the first try—but the competition is what makes this exciting.
Testing Anthropic's latest model against its predecessor and the competition
The AI race continues to heat up. Hot on the heels of my GPT-5 vs Claude Opus 4.1 comparison, Anthropic has released Claude Sonnet 4.5, and naturally, I had to put it through the same Pacman challenge.
The usual disclaimer: This is not a scientific comparison. It's just a fun way to explore what these models can do, and honestly, I love watching this competition unfold.
The Challenge Revisited
Same rules as before: create a functional Pacman game from a single prompt, no follow-ups, no corrections. Just pure first-attempt execution.
The Results
Here's where things got interesting. Unlike its siblings Opus 4.1 and GPT-5, which both nailed it on the first try, Sonnet 4.5 had some hiccups:
What went wrong:
- Pacman spawned inside a wall
- Gameplay speed was extremely fast, making it nearly unplayable
- Everything else functioned correctly
What went right:
- Core game mechanics were implemented
- Visual elements rendered properly
- The overall structure was solid
Comparison at a Glance
| Model | First Attempt Success | Notable Issues |
|---|---|---|
| GPT-5 | ✅ Yes | None |
| Claude Opus 4.1 | ✅ Yes | None |
| Claude Sonnet 4.5 | ⚠️ Partial | Spawn position, speed balance |
What This Tells Us
Sonnet 4.5 is positioned as Anthropic's most efficient everyday model, optimized for speed and routine tasks. This Pacman challenge, while far from scientific, suggests there might be trade-offs between efficiency and the meticulous attention to detail that Opus 4.1 demonstrated.
That said, these are minor issues that would be trivial to fix with a single follow-up prompt. The model clearly understood the assignment and delivered a mostly functional game.
The Bigger Picture
What excites me most isn't declaring a winner, but watching how quickly this technology is evolving. We now have multiple AI models that can turn a simple request into playable games. The differences between them are becoming more nuanced, more about style and approach than capability.
The competition between OpenAI and Anthropic is pushing both companies to innovate faster, and we all benefit from that.
About the Author
Kemal Esensoy
Kemal Esensoy, founder of Wunderlandmedia, started his journey as a freelance web developer and designer. He conducted web design courses with over 3,000 students. Today, he leads an award-winning full-stack agency specializing in web development, SEO, and digital marketing.