GILD Review

No subscription. Standard pay-as-you-go bills each model per million tokens through Google Cloud. Gemini 4 Argon lists at US$4 input and US$20 output per million tokens on the global endpoint; through Nov 30, 2026 a promotion returns 50% of net Argon spend as credits, an effective US$2 and US$10; Claude Opus 5.5 is US$4 and US$20. Provisioned Throughput reserves capacity in GSUs, from US$7.14 per GSU-hour on a 1-week term to about US$2.74 on a 1-year term.

Model choice, input and output tokens, endpoint type and cache reads set the bill. Non-global endpoints cost 10% more for Gemini 3 and later models. Grounding with Google Search adds its own per-request charge. Promotional rates on Gemini 4 Argon and the Gemini 3.x Flash models end on Nov 30 and Dec 31, 2026. If a request exceeds the remaining Provisioned Throughput quota, the whole request is processed on demand by default and billed at the pay-as-you-go rate.

Agent Platform runs in Google Cloud projects. Requests choose the global endpoint, which covers the entire world with higher availability than single regions, the US or EU multi-region endpoints, or a regional endpoint such as us-central. Google notes that endpoints don't guarantee data residency or in-region ML processing.

Standard PayGo Tier 1 sets 500,000 organization-level TPM for Gemini Pro and Argon models. Model Garden lets you choose models managed by Google, partners, or you. Workforce Identity Federation (GA) provides single sign-on, except Vertex AI Pipelines in the console and Colab Enterprise. Google lists Generative AI on Agent Platform for ISO 27001. Admin Activity audit logs are always on, Data Access audit logs off by default. Models as a Service retains customer data for limited periods in six areas.