The Robotic Tortoise and the Robotic Hare (tomtunguz.com)

In a recent experiment comparing two AI models, a local version named Qwen 35B and Claude Opus 4.5, both were tasked with building a payment app on Stripe’s new Tempo blockchain. Despite Opus 4.5 being significantly more advanced—around 20% smarter and 50 times larger than Qwen—the local model completed the task in just 2 minutes, whereas Claude took over 6 minutes. This outcome highlights an interesting paradox in AI development: faster response times can yield quicker iterations and better overall results in practical applications. The study emphasizes the importance of efficiency over sheer capability in AI/ML workflows, particularly for routine tasks. Faster models like Qwen allow for tighter feedback loops, enabling more revisions before meetings or decisions are made. While more complex coding projects might benefit from deeper, slower analyses, for everyday assignments, speed can lead to superior outputs. This critical insight suggests that the smartest AI isn’t always the most effective, urging the AI community to rethink priorities in model development and application strategies.