Gemini 4 Argon is live in n8n at $2/$10 per million tokens. Budget for $4/$20, because that price ends November 30

Google's newest frontier model is now in the API and in n8n, but the launch price has an expiry date that belongs in every client quote.

What happened

Google made Gemini 4 Argon globally available to developers and enterprises through the API in AI Studio and the Gemini platform. The introductory price is $2 per million input tokens and $10 per million output tokens, with cached input tokens 95% off. Google says that after the introductory period ends on November 30, 2026, the price becomes $4 per million input and $20 per million output. Google also describes a 1M unified input-output token limit via the Gemini API, plus a new continuation protocol for long tasks.

In the Gemini app, Ars Technica reports Argon is only for paid AI Pro and AI Ultra subscribers, and that an automated misalignment monitor can stop processing if it detects an unintended workflow.

n8n says Argon is available from launch as a model for its AI Agent and in the standalone Google Gemini node. On n8n Cloud it can run through Gateway credits, so no separate provider account is needed. On self-hosted, the n8n Assistant has no native Gemini option yet, so you reach it through OpenRouter or Google's OpenAI-compatible endpoint.

My take

The model is not the decision. The price curve is. If you quote a client a monthly AI cost this week at $2/$10, that number is wrong in seven weeks. I would model every workflow at $4/$20 and treat the launch window as a discount, not a baseline.

Second, place Argon where judgment matters: the step that reads a messy support email, picks the CRM record and decides priority. Leave field mapping, routing by rules and formatting on plain nodes or a cheaper model. One expensive step inside a predictable workflow is easy to budget. An expensive model on every step is not.

Third, a shared input and output budget changes long runs. If an agent writes long reports, plan for output eating into the room you thought you had for context.

Finally, n8n's MCP flow builds a workflow from your IDE but stops before publishing and asks first. Keep that habit for anything a model builds for you.