Datadog announced on 15 October 2026 Scout 1.0, which it calls its first AI model specialised for investigating production issues. Scout powers Bits Investigation, Datadog’s AI SRE. The company says Datadog engineering and a small group of customers are already using it in a closed preview, with a sign-up path before general availability.
This is a preview launch, not GA. The performance and cost comparisons below are Datadog’s own claims on internal and selected external benchmarks.
What Datadog says Scout is for
Bits Investigation is framed as an always-on first responder: form hypotheses, query telemetry, and decide whether evidence supports a conclusion before an engineer steps in. Those loops mean repeated model calls, so Datadog says it trained Scout to cut inference cost enough to run investigations more often and earlier in an incident.
Datadog says Scout 1.0 is built on GLM-5.3 Flash and post-trained with in-house reinforcement learning in investigation simulations drawn from historical incidents and telemetry that do not contain customer data. A technical report PDF is linked from the blog for fuller methods detail.
Claims on AlertEval (read the footnote)
On AlertEval, Datadog’s internal human-expert-annotated investigation benchmark, the company claims:
- Scout 1.0 medium matches GPT-6.1 Sol high and Claude Opus 5.5 high at 3–8× lower estimated inference cost
- it performs comparably to Fable medium and Astra medium at 12× and 19× lower cost
- with test-time scaling, Scout 1.0 xhigh matches Fable xhigh at 4× lower cost
Datadog’s footnote is essential: scores measure agreement with expert reference explanations on a 0–100 scale; they are not percentages of incidents solved. Costs are estimated inference costs per investigation, relative to Scout 1.0 medium.
Do not restate those numbers as an independent frontier ranking or as a solve-rate percentage.
SREGym note
Datadog also says it evaluated Scout on SREGym, a third-party benchmark, claiming Scout 1.0 medium improves on the GLM-5.3-Flash base at more than 4× lower cost, and that Scout 1.0 high is competitive with the best models there while keeping at least 2× cost savings (and about 10× versus Astra/Fable-class models). Again, treat this as the company’s reading of the benchmark.
Preview status and roadmap colour
Preview customers use Bits Investigation powered by Scout on production alerts. Looking ahead, Datadog says it wants wider failure simulations, still-lower inference cost, and eventually help with autonomous remediation — moving Bits from first responder toward resolving issues. That remediation path is stated as a plan, not a shipping Scout 1.0 feature in the announcement.
Context from earlier TSN coverage
Agent runtimes and sandboxes are the other half of “AI that operates production.” TSN recently covered Cline cloud sessions and Perplexity’s open-source SPACE sandbox. Scout is Datadog’s bet that a specialised, cheaper investigation model belongs inside the observability console, not only in a general chat agent.
Why it matters
If Bits is to sit on every alert, unit economics matter as much as raw model quality. Datadog is selling a specialised SRE model on a price–performance curve of its own making. Readers should keep the 0–100 agreement definition in view whenever those curves are quoted.
What’s still unknown
- When Scout / Bits Investigation leave closed preview for GA
- How AlertEval maps to customer environments Datadog does not control
- Independent replications of the cost multiples
What to watch
Watch Datadog’s preview feedback notes and whether SREGym or AlertEval numbers are updated with third-party commentary.
What this means
- Scout 1.0 powers Bits Investigation and is in closed preview.
- Built on GLM-5.3 Flash with Datadog’s RL sims (company description).
- Benchmark scores and cost multiples are Datadog’s; AlertEval scores are agreement 0–100, not percent solved.
Sources
- Datadog, “Introducing Scout 1.0: A new SRE model powering Bits AI…”, 15 October 2026
- Datadog Scout technical report (PDF)
Related stories
- Cline v4.1.25 Adds Cloud Sessions That Keep Running After You Close the Editor
- Perplexity Open-Sources SPACE, Its Sandbox Platform for Agentic Code Execution
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
