Scout 1.0 Datadog's first specialized AI model for investigating production incidents and powering Bits Investigation, its AI SRE agent. Scout 1.0 was post-trained from GLM-5.3 Flash using reinforcement learning on realistic simulations of real-world incidents and production telemetry, matching frontier models like Claude Opus and GPT variants on internal AlertEval and public SREGym benchmarks at 3-19x lower cost.
Visit productFeatured media
Loading media…
View on XCosigns Received
“It was a blast to get to contribute to this effort! Working with realistic incident scenarios and models in the hundreds of billions of params, this is serious post training scale Congrats to the team, and we'll have more coming on this front”
“We post-trained GLM 5.3 Flash to investigate prod incidents, matching frontier models at a fraction of the cost: - Scout 1.0 (med) matches Opus 5.5 & GPT-6.1 Sol (high) at 3–8x lower cost - Scout 1.0 (xhigh) matches Fable 5.1 & beats GPT-6 Astra (xhigh) at 4x lower cost”
“#Datadog が開発した障害調査に特化したAIモデルが公表されました！ ベンチでは effort medium が GPT-6.1 Sol high や Opus 5.5 highと同等水準の調査品質を1/3〜1/8、xhighがFable xhighの約1/4のコストで結論づけれたとのことです 🤑 障害調査という品質と速度が求められる業務を低コストで実現できて人類の救いですね 👼 Bits Investigation が実現する未来が益々楽しみです！”
“Exciting news by our research team. Scout 1.0 Medium matches Claude Opus 5.5 High and GPT-6.1 Sol High at a fraction of the cost.”
“Two years ago, when we started building our SRE agent, we bet heavily on evals. We hand-labelled thousands of production incidents and built the infrastructure needed for our engineers to run this golden dataset every day as they iterated on the agent. That work underpins how we evaluate Scout 1.0. Specialized models start with knowing what good looks like in your domain.”
“So excited about this one We (@datadoghq) post-trained our own model for incident investigation Scout matches Opus 5.5 and GPT-6.1 on our AlertEval benchmark at a fraction of the cost. already running with a few customers and we'll progressively serve more and more customers who can sign up to the private beta. this will let us give our customers always on agents detecting problems, finding the root cause and solving them completely autonomously Congrats to our research team, it's been awesome seeing them execute!”