Datadog launches Scout 1.0, an RL-tuned GLM-5.3 Flash for incident investigation: matches Opus 5.5 and GPT-6.1 Sol on its own AlertEval at 3-8x lower cost; closed preview

Why it matters

【Key points】

On October 15 Datadog released Scout 1.0, its first in-house model, built to investigate production incidents and power its AI SRE agent, Bits Investigation. Scout starts from Z.ai's open-weight GLM-5.3 Flash and is post-trained with reinforcement learning in investigation simulations built from real historical incidents and alert telemetry (Datadog says no customer data), rewarded for finding the cause and backing it with evidence.

Datadog's numbers: on its internal, expert-annotated AlertEval, Scout 1.0 medium matches GPT-6.1 Sol high and Claude Opus 5.5 high at 3–8x lower estimated inference cost, and Fable and Astra medium at 12x and 19x lower cost; xhigh matches Claude Fable 5.1 xhigh at 4x lower cost. On third-party SREGym, medium clearly beats the base model and surpasses GLM-5.3 at under a quarter of the cost. It is in use internally and in a closed customer preview with sign-up; no GA date or pricing.

【Verifiable】

Blog: https://www.datadoghq.com/blog/ai/introducing-scout-sre-specialized-model-bits-ai/

Technical report: https://web-assets.dd-static.net/42588/1792071077-datadog-scout_technical_report.pdf

@datadoghq: https://x.com/datadoghq/status/2110750909152034912

【Our read】

1. Another case of open base plus vertical RL (after Paper Instruments' Ultramarine, e/877, also on GLM-5.3 Flash): companies with real environments and data are closing in on frontier models on narrow tasks with small ones.

2. The headline results are on Datadog's own AlertEval, scored as agreement with expert explanations (0–100), not incidents solved; SREGym is the better outside check.

3. For users the point is cost: investigations ten times cheaper make it practical to run one on every alert.