What AI says about CompassJudger-1

CompassJudger is an open-source project that develops CompassJudger-2, a generalist judge model designed to evaluate large language models across multiple tasks and domains, using a task-driven data curation strategy and verifiable rewards. It employs rejection sampling to guide intrinsic critical reasoning and uses a margin policy gradient loss to improve judgment performance, with a 7B model achieving competitive results against larger models like DeepSeek-V3 and Qwen3-235B-A22B. The project also introduces JudgerBenchV2, a benchmark suite for cross-domain judgment accuracy and rank consistency to standardize judge-model evaluation.