The Claude Code Opus 4.6 Performance Tracker aims to identify significant performance degradations in Claude Code's capabilities on software engineering tasks. It updates daily and benchmarks against a carefully selected subset of SWE-Bench-Pro. The tracker conducts statistical tests to pinpoint any drops in performance, ensuring that the evaluations reflect real-world user experiences by running benchmarks directly in the Claude Code CLI without introducing custom setups. The tracker provides several metrics to monitor performance over time. Daily trends include pass rates and confidence intervals, while weekly trends aggregate data to offer a clearer picture of performance stability. Currently, baseline data is being collected, and performance deltas will be available once enough data is gathered. Each day, evaluations are conducted on 50 test instances, which accounts for daily variability. The methodology employs Bernoulli random variables to compute pass rates and their confidence intervals, enabling the identification of statistically significant changes. Additional metrics track resource usage, including daily input and output tokens, API costs, and average runtimes. These insights help users understand the cost and efficiency of using Claude Code. The initiative is particularly relevant given Anthropic's past postmortem on performance issues, as it aims to provide a reliable resource for spotting future degradations in the model's performance.