Algobench
Supercharging algorithm development
Algobench is a platform for evolving and benchmarking optimization algorithms in real-world, production-grade settings. Inspired by DeepMind's research on FunSearch and AlphaEvolve, Algobench brings the principles of automated discovery, evaluation, and improvement of algorithms into production.
Key Features
- Benchmark and evolve your Python optimization algorithms
- Integrate with your existing stack via one simple decorator
- Store and analyze problem instances and algorithm performance
- Systematically improve solution quality without manual intervention
Perfect For
- Vehicle routing optimization
- Production scheduling
- Portfolio optimization
- Supply chain problems
Getting Started
- Contact us at to get access.
- Install the SDK from GitHub to integrate with your codebase
- Run your algorithm and visit Problems to see the results
Case Study: Zalando Order Batching
Large-scale logistics involves a range of high-impact optimization problems. At Zalando, the Logistics Algorithms department focuses on optimizing outbound processes: picking items from storage, merging them into orders, and preparing them for shipment. A core challenge is determining which customer orders to process next — in other words, deciding which items a warehouse worker should collect on a picking cart (essentially a large-scale version of a supermarket cart).
Recently, Zalando's Logistics Algorithms team published a technical report on the order batching and routing problem, including code and problem definitions. This made it an ideal candidate for Algobench. We adapted the problem to run on Algobench and benchmarked it. Compared to the baseline algorithm, which achieved a score of 41878, Algobench reached 36716, improving performance by 12%. At Zalando's scale, this improvement translates to significant, multi-million-euro cost savings across its fulfillment network.