Datadog has introduced Scout 1.0, its first model built specifically for site reliability and operations troubleshooting, using Zhipu AI’s open-source GLM-5.3 Flash as the base model. The company said Scout is designed to investigate production incidents by querying logs, monitoring metrics, and service call traces, then narrowing down root causes step by step.
According to Datadog’s self-reported results, Scout completed 80 diagnosis-and-remediation tasks on the public SREGym benchmark, which contains 90 failure investigation tasks. That matched GPT-6 Astra xhigh and trailed GPT-6.1 Sol xhigh by 2 tasks. The original GLM-5.3 Flash completed 66 tasks on the same benchmark.
Datadog also said that in its internal AlertEval evaluation, Scout at the medium reasoning tier delivered diagnostic quality close to GPT-6.1 Sol high and Claude Opus 5.5 high, while carrying an estimated inference cost 3x to 8x lower. Compared with GPT-6 Astra medium, the cost was about 19x lower. Scout has already been integrated into Datadog’s Bits Investigation and is currently available only to a small group of customers in a closed preview.
Datadog has launched Scout 1.0, its first model dedicated to operations and incident troubleshooting, built on Zhipu AI’s open-source GLM-5.3 Flash.
The model is designed to investigate failures in live systems. It can query logs, monitoring metrics, and service call traces, then work through the evidence to identify the root cause of an incident. Datadog said it trained Scout in simulated environments built from real incidents and monitoring data, using reinforcement learning so the model could learn to discard misleading signals and support its diagnostic conclusions with evidence.
Benchmark and internal evaluation results
In Datadog’s self-reported testing on the public SREGym benchmark, which includes 90 failure investigation tasks, Scout at its highest reasoning setting completed 80 diagnosis-and-repair tasks. That score matched GPT-6 Astra xhigh and came in 2 tasks behind GPT-6.1 Sol xhigh. The original GLM-5.3 Flash completed 66 tasks.
In Datadog’s internal AlertEval evaluation, the company said Scout at the medium setting produced diagnostic quality close to GPT-6.1 Sol high and Claude Opus 5.5 high. Datadog estimated inference costs were 3x to 8x lower, and about 19x lower than GPT-6 Astra medium.
Current availability
Scout has been integrated into Datadog’s Bits Investigation and is currently in closed preview for a limited number of customers. For now, the model is focused on identifying the cause of incidents and suggesting recovery actions. Automatic execution of fixes remains a later development goal.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
