We made Grok 4.5, GPT-5.5, and Claude build the same apps
tryai.dev
July 8, 2026
6 min read
🔥🔥🔥🔥🔥
56/100
Summary
Grok 4.5 has launched, described as xAI's smartest model yet and trained alongside Cursor for coding and agentic tasks. A build-off was conducted where Grok 4.5, GPT-5.5, Claude Opus 4.8, and Fable 5 created the same interactive apps, with results measured for latency and cost.
Key Takeaways
- Grok 4.5 was launched as xAI's smartest model, specifically trained for coding and agentic tasks.
- In a build-off, Claude Opus 4.8 and Claude Fable 5 successfully created a 3D Rubik's Cube app on the first attempt, while Grok 4.5 required a retry.
- GPT-5.5 produced the most visually appealing particle gravity sandbox, winning based on aesthetic appeal.
- All models successfully created a playable Breakout game on the first try, resulting in a tie for that task.
Community Sentiment
Positives
- Impressive, especially the number of models used for the comparison — it really showcases the landscape of AI capabilities.
- This arena of 52 apps across 21 models is a goldmine for anyone interested in real-world AI performance metrics.
Concerns
- Comparisons are completely subjective and the problem space is too simplistic for today's AI — it's like a kid trying to do science.
- The benchmarks seem cherry-picked, ignoring the nuances of what makes a model succeed or fail — not a reliable evaluation.
- I worry that newer models will be heavily restricted, which could mess up actual benchmarks and lead to misleading results.
