Love seeing projects like this. The performance benchmarks are nice to see. Have you done any benchmarks against approaches like OpenAI's Symphony for things like token usage or task completion?