During August 2025, Yue developed a Performance Benchmark Snapshot Table for the MiroMindAI/MiroThinker repository, enabling direct comparison of token processing speeds across models on both MacBook M4 Pro and RTX 5070 hardware. Yue’s approach centered on establishing a reliable benchmarking workflow, aggregating performance data, and documenting results using Markdown. This work provided actionable insights for optimizing model efficiency and informed decisions on performance-cost trade-offs. While no major bugs were addressed, Yue focused on maintaining and refining benchmarking scripts. The depth of the contribution lay in improving visibility into model throughput, supporting data-driven prioritization for future optimization efforts.