OpenCost 1.121.0: First-of-a-kind Kubernetes inference cost tracking

Sicherheitsmeldung mit Schweregrad noch nicht bewertet. Technische Details im Tab „Originaltext“; empfohlene Schritte in der Checkliste.

Originaltext der Quelle (nicht übersetzt) – die deutsche Zusammenfassung findest du im Tab „Kurzfassung".

Your GPU bill is rising. Your models are serving billions of tokens. Yet one question remains unanswered: what does each token actually cost?

This is not a hypothetical problem. Platform teams today operate in a fog regarding AI inference costs—they see infrastructure spend and track token throughput, but the connection between those numbers is invisible. Without per-model and per-token costs derived from actual resource consumption, every decision becomes a gamble.

Is self-hosting cheaper than using a SaaS API? You’re guessing.

Which model is actually cost-efficient at your traffic levels? The data doesn’t exist.

Which team’s agent workload is consuming your AI budget? Nobody knows.

The result is a large monthly bill with no clear explanation, while executives ask hard questions about AI return on investment (ROI) that you cannot answer with numbers.

Cost ≠ Price: SaaS providers may price below cost to gain market share or well above cost for premium models. When comparing self-hosting costs to SaaS, keep this distinction in mind. An enterprise’s cost for SaaS inference is the provider’s price.

To close this gap, we are integrating OpenCost, a CNCF incubating project, with llm-d, a CNCF sandbox project for distributed LLM inference on Kubernetes. This post explains how the integration works, what metrics it produces, and how platform teams can use them to make data-driven decisions.

Note: vLLM users who do not use llm-d can also benefit from these capabilities, since the core metrics used to calculate model and per-token costs come from vLLM.

GPUs are just CPUs …

Vollständiger Originalartikel ↗

• Betroffenheit im eigenen Stack prüfen: Versionen/Komponenten abgleichen.
• Originalquelle / Hersteller-Advisory lesen: https://www.cncf.io/blog/2026/08/05/opencost-1-121-0-first-of-a-kind-kubernetes-inference-cost-tracking/
• Verfügbaren Patch oder Workaround einspielen und dokumentieren.