Frontier SRE agent trained by Datadog

Scout 1.0 is a production-focused SRE model developed to power automated incident investigations within Bits Investigation, Datadog’s AI SRE. It’s specialized for diagnosing production issues and was engineered to drive down inference cost so investigations can run far more frequently and earlier in an incident lifecycle. The team trained Scout with in-house distributed training and efficiency techniques, and built test-time scaling into the deployment strategy to let smaller, cheaper variants match higher-cost frontier behavior when needed.

Benchmarks report substantial price-performance gains versus general-purpose frontier models: Scout 1.0 medium matches GPT-6.1 Sol high and Claude Opus 5.5 high at an estimated 3-8× lower inference cost, and is comparable to Fable medium and Astra medium at roughly 12× and 19× lower cost, respectively. With test-time scaling, Scout xhigh reaches Fable xhigh performance at about 4× lower cost. Those concrete cost-performance trade-offs are the central finding: comparable diagnostic capability for SRE scenarios at a fraction of the operational cost, enabling more frequent automated investigation runs and lower overall incident-resolution expense.