SWE-bench Verified

TokenCast: Predicting Token Usage During LLM Agent Execution

TokenCast predicts LLM agent token usage from call counts, context growth, and cost residuals, improving budge…