news

Dynatrace closes its Arize acquisition, putting a widely used LLM-judge stack under an observability vendor

The deal completed on October 1 leaves Phoenix open source and Arize AX standalone, with no integration timetable and two different reported price tags.

By Daniel Ochoa · 2 October 2026 · Updated 4 October 2026

The tooling many teams use to decide whether a model output was any good has changed hands. Dynatrace announced on October 1 that it completed its acquisition of Arize, bringing Phoenix, Arize AX and the OpenInference instrumentation conventions into an enterprise observability company that sells to site reliability and platform engineering teams.

What actually closed, and what stays where it is

Dynatrace frames the combination as tracing, evaluation and experimentation from Arize sitting alongside its own monitoring of applications, infrastructure, user experiences and business processes. The practical detail for anyone with Phoenix in a pipeline is continuity: Phoenix continues as an open-source project, OpenInference continues as an OpenTelemetry-compatible project, and Arize AX remains available as its own managed platform. Dynatrace says the companies are now starting joint product and roadmap planning, with priorities informed by customers and developers, and it has not published an integration schedule. Intent to acquire was announced in August.

The reported price does not agree with itself

Dynatrace's completion post does not state terms. Coverage from August of the original announcement described roughly 815 million dollars in cash plus replacement equity awards for Arize employees, with closing expected late in the quarter or early in the next fiscal quarter. At least one deal tracker reported the completed transaction at 915 million dollars all cash. Those two figures can be reconciled if the larger number includes equity awards, but neither company has published a reconciliation, so the correct answer is that the number is not settled publicly.

Why a media-scoring site cares who owns Phoenix

Phoenix is one of the more honest pieces of judging infrastructure in circulation, because it treats the judge as something to be audited rather than trusted. Its evaluators extract structured judgments through tool calling instead of parsing freeform text, evaluator runs are themselves traced so the input data, the exact prompt sent to the judge, the judge's reasoning, the final scores and timing are all recorded, and the documentation pushes users to benchmark a custom judge against human-annotated ground truth before relying on it. That is close to the disclosure standard a public leaderboard should meet and very rarely does.

Image judging depends on which documentation page you open

Support for scoring visual output is less clear than the acquisition coverage implies. The Arize AX guidance on custom judge templates tells users to map an image or audio reference alongside the text output for multimodal applications, and points to a cookbook where a judge labels structured extraction from a source image as grounded, not grounded, or needs review. An older Phoenix cookbook page still carries a line saying the evals library does not yet support multimodal inputs such as images. One of those pages is stale, and neither Dynatrace nor Arize addressed media evaluation in the completion announcement, which is written entirely around agents, retrieval, tool use, cost and reliability.

What to watch next

Two things decide whether this matters to anyone grading generated video or images. The first is whether Phoenix's multimodal evaluation path gets first-class support and documentation under new ownership, or stays a cookbook footnote while engineering attention goes to agent traces, which is where the enterprise revenue is. The second is concentration risk: a default open-source eval tool owned by a public monitoring vendor is still open source, and the stated commitment is explicit, but the roadmap now answers to a different set of customers than the AI engineers who adopted it.

Sources: Dynatrace, Arize AX docs, Arize Phoenix docs, DevOps.com.