Compute:Arena

ComputeArena offers a comprehensive platform for community-submitted performance benchmarks of local models running on various hardware configurations. Key features include: • Support for any hardware and runtime environment • Benchmarks across different quantisation levels • Detailed throughput metrics like decode and prefill tokens per second • Reports sourced from real community hardware submissions • Interactive terminal-based installation and usage on macOS and Linux This platform enables users to explore in-depth performance data sorted by model, quantisation, and chipset, helping to identify optimal configurations for edge devices. By aggregating data from contributors worldwide, it provides transparency and real-world insight into how models perform on diverse silicon architectures. The installation process is simple and requires no elevated permissions, making it accessible for developers and researchers. Designed for developers, researchers, and hardware enthusiasts interested in evaluating and comparing local AI model inference speeds on edge devices, ComputeArena bridges the gap between theoretical performance and actual throughput. It supports informed decision-making for deploying models efficiently on target hardware. Whether for benchmarking new chips or tuning model deployment strategies, this tool offers a community-driven, data-backed resource.