#LLM judge04/10/2025
TUMIX: Google’s Multi-Agent Tool Mixture That Boosts Hard-Reasoning Accuracy
'TUMIX ensembles diverse tool-using agents that share notes and use an LLM judge for early stopping, improving accuracy on hard reasoning benchmarks while cutting inference cost.'