SRE-1: 27B model for Production Investigations

resolve.ai49 min read

Researchers introduce SRE-1, a 27B model trained with reinforcement learning to find the root cause of production alerts. The model dominates the Pareto frontier of pass rate against cost on difficult live production cases, raising single-investigation pass rates from 6.9% to 38.8%. An ensemble of five SRE-1 investigations matches the strongest frontier model measured while costing 37% less. Open article →