BENCHMARKS Function Calling Harness Pushes Qwen From 6.75 Percent to 100 Percent Success on Complex Schemas 8/10 4 min read 1 month ago