Datadog launches an operations model based on Zhipu GLM-5.3 Flash, Scout 1.0, in closed preview

Article

Datadog Zhipu GLM training and operations AI：90 fault evaluation tasks 80 tasks, and GPT-6 Astra tied

Beating AI breaking news, U.S. monitoring software company Datadog has launched its first dedicated operations model, Scout 1.0, built on Zhipu's open-source GLM-5.3 Flash.

It is specifically responsible for troubleshooting online system failures. It can query logs, monitoring metrics, and service call chains, and progressively identify the root cause of failures.Datadog built a simulated environment using real incidents and monitoring data, then trained the model through reinforcement learning so that it could learn to eliminate misleading clues and support diagnostic conclusions with evidence.

Among the 90 fault-troubleshooting tasks in the public SREGym benchmark,Datadog reported that Scout at its highest reasoning level completed 80 diagnoses and repairs, tying GPT-6 Astra xhigh and falling short of GPT-6.1 Sol xhigh by just 2 tasks. The original GLM-5.3 Flash completed 66 tasks.

In Datadog's internal AlertEval evaluation, the diagnostic quality of Scout's mid-level setting was close to that of GPT-6.1 Sol high and Claude Opus 5.5 high, while its estimated reasoning cost was only 3 to 8 times lower. Compared with GPT-6 Astra medium, the cost was approximately 19 times lower.

Scout has been integrated into Datadog's Bits Investigation and is currently available only to a small number of customers in a closed preview. At this stage, it is mainly responsible for determining the cause of failures and proposing recovery recommendations; automatically executing repairs remains a future research and development goal.

Original link https://m.theblockbeats.info/flash/372284

Key points

Datadog has launched its first dedicated operations model, Scout 1.0, built on Zhipu's open-source GLM-5.3 Flash。
Scout can query logs, monitoring metrics, and service call chains to progressively troubleshoot the root causes of online system failures.
Datadog claims that Scout at its highest reasoning level completed 90 fault tasks in SREGym and completed 80 diagnoses and repairs.
Datadog claims that Scout tied GPT-6 Astra xhigh, falling short of GPT-6.1 Sol xhigh by 2 tasks; the original GLM-5.3 Flash completed 66 tasks.
Datadog claims that Scout's mid-level setting had diagnostic quality in the internal AlertEval evaluation close to that of GPT-6.1 Sol high and Claude Opus 5.5 high, with estimated reasoning costs 3 to 8 times lower.
Scout has been integrated into Datadog's Bits Investigation and is currently available only to a small number of customers in a closed preview.
At this stage, Scout is mainly responsible for determining the cause of failures and proposing recovery recommendations; automatically executing repairs remains a future research and development goal.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

Scout is an operations AI model used to troubleshoot online failures. It examines logs, monitoring metrics, and service call chains, then uses evidence to identify possible causes of failures and propose recovery recommendations. The performance and cost data in the article mainly come from Datadog self-evaluations or internal evaluations.

Why it matters to readers

This shows that operations AI is attempting to enter real-world enterprise fault-troubleshooting workflows. Scout has been integrated into Datadog's product, but it is currently still in a closed preview for a small number of customers, and automatic repair has not yet been implemented.

It can be understood as an AI assistant that helps operations personnel troubleshoot failures: it currently mainly provides analysis and recommendations, rather than being a fully automated repair tool.

Risks and unknowns

The core performance and cost results in SREGym and AlertEval have not been shown in the evidence to have been independently verified.
Scout is currently available only to a small number of customers in a closed preview; the status of public testing, formal commercialization, and customer adoption is unknown.
Automated remediation remains a future research and development objective; its actual reliability and security are unknown.
Zhipu and GLM-5.3 Flash have achieved clear commercial returns; the evidence does not state whether this is the case.

Datadog launches an operations model based on Zhipu GLM-5.3 Flash, Scout 1.0, in closed preview

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center