What is Weco?

You've heard of vibe coding, but have you heard of metric-driven coding? Weco is an AI Research Engineer that automates trial-and-error driven code optimization by leveraging our agentic system powered by AIDE. We iteratively refine your code against any metric (such as speed, accuracy, or cost) and deliver a better-performing solution.

Our core engine AIDE has been shown to outperform state-of-the-art code generation systems by a wide margin as shown in OpenAI's MLE-bench and METR's RE-bench to name a few. It's even shown to outperform human engineers and scientists on some tasks. Sakana AI's AI Scientist v2 pushes the boundary of automating scientific discovery using AIDE as its foundation. AIDE continues to lead the pack in metric-driven generation of solutions and is used by leading technology companies, frontier AI research labs, and top universities.

Here's a sneak peak of what Weco can do (keep your eye on the Best Solution panel):

Why Weco?

• user-centric: Weco enables you to optimize your code using your own metrics, your own data and your own evaluation pipeline.
• flexible: Weco can be used to optimize code in any language. From CPUs to GPUs to TPUs to even Apple Silicon. Whether you're using PyTorch, TensorFlow, or even just plain Python, Weco can help you optimize your code.
• scalable: Weco is designed to scale to complex optimizations tasks across long time horizons as shown below in an independent evaluation by METR.