New AI Models Race to Spot Centipedes — Which One Wins?
Faster AI could mean real-time help in your pocket.
A few weeks ago, several new AI models designed for making quick decisions appeared online. A curious tester decided to compare four of them—Laya, Liquid's d1, Cloudflare's Clef-Flash, and Interfaze's Lev—on the same powerful graphics card (RTX 4090) to see which one was fastest. The task: read nine Wikipedia articles about centipedes (9,534 words) and flag every word that names a centipede. Each model processed one word at a time, and the tester measured how long each took.
The results showed big differences. Laya was the speed champion, handling each word in just 3.9 milliseconds and finishing all words in 32 seconds. d1 was next at 6.0 milliseconds, followed by Clef-Flash at 24.4 milliseconds, and Lev at 51.0 milliseconds—13 times slower than Laya. However, speed isn't everything. Lev caught the most centipede names (83% accuracy), while Laya caught 70%. The high overall accuracy numbers (around 97%) are misleading because simply saying 'no' to most words counts as correct, since most words aren't centipede names.
The tester noted that Lev's slowness was partly because it runs in its own PyTorch server instead of the optimized llama.cpp engine used by the others. Also, Lev's performance was sensitive to the computer's processor. For most practical uses, Laya's speed and ease of customization make it a top pick. This test highlights that even on the same hardware, AI models can vary dramatically in speed and accuracy—factors that matter when you need instant responses in apps like voice assistants or real-time translation.
For everyday users, this means that the AI behind your favorite apps could become much faster and more capable soon. Faster models like Laya could enable real-time language processing on your phone, while more accurate models like Lev might be better for tasks where precision is critical, like medical coding. The competition among these small, efficient models is driving rapid improvements that will likely lead to smarter, quicker AI tools in your daily life.
- Laya was the fastest model, processing 7,980 words in 32 seconds—about 3.9 milliseconds per word.
- Lev was 13 times slower but caught the most centipede names (83% accuracy).
- Speed differences matter for real-time apps like voice assistants or live translation.
Why It Matters
Faster AI models could soon power real-time apps on your phone, making them more responsive and useful.