Where Gemini billing actually lives
- AI Studio keys: the simple path. Usage in AI Studio, billing on the attached Google account, a free rate limited tier for prototyping per AI API free tiers.
- Vertex AI: the enterprise path. Spend lands in Cloud billing with your other GCP costs, sliceable by project and label, exportable to BigQuery. Genuinely strong infrastructure accounting, and still customer blind.
Pricing has more dimensions than the other providers: model tiers (pro, flash, flash lite), context length thresholds on some pro models, media inputs priced separately from text. Current text rates per million run from about $0.30 and $2.50 on gemini-3.5-flash-lite to $1.50 and $7.50 on gemini-3.6-flash, live and watched at useweckr.com/pricing.json.
The per user pattern for Gemini
Every response includes usage metadata with prompt and completion token counts. The manual version: read it, store it with your user id, price it. The two line version with Weckr, which detects the Gemini client natively:
import { GoogleGenAI } from '@google/genai';
import { Weckr } from '@weckr/sdk';
const gemini = new GoogleGenAI({ apiKey: process.env.GEMINI_API_KEY! });
const wk = new Weckr({ apiKey: process.env.WECKR_API_KEY!, plans: { free: 0, pro: 29 } });
const result = await wk.chat(gemini, {
model: 'gemini-3.6-flash',
messages,
userId: user.id,
feature: 'doc-analysis',
plan: user.plan,
});The call returns Gemini’s response unchanged; cost is recomputed server side at current rates and logged per user, per feature, against the user’s plan price. From there, margins, caps, and runaway detection work identically across your providers, one dashboard whether the call went to Gemini, OpenAI, or Claude, which matters because tokenizers and rates differ enough that cross provider comparison by hand is its own project, per tokens to dollars.
Gemini specific cost notes
- Watch the flash generation drift. The 3.x flash line prices above the 2.5 flash era, a quiet increase documented in LLM price history. Generation upgrades are not always price cuts.
- Mind context thresholds on pro. Long context requests can cross into higher rate territory on pro tier models; if you do RAG with big contexts, measure rather than assume.
- Same provider downgrades exist. Flash to flash lite is a clean cap downgrade path, the mechanics of model downgrade on budget.
FAQ
Where do I see my Gemini API costs?
It depends how you access Gemini. Through AI Studio API keys, usage appears in AI Studio and billing runs through the Google account attached to the key. Through Vertex AI, spend lands in Google Cloud billing with the rest of your GCP costs, sliceable by project and label. Either way, the reporting grain is your account and project, never your end users.
How is Gemini billing different from OpenAI or Anthropic billing?
Structurally the same per token model, wrapped in Google Cloud machinery. Rates are per million tokens with cached input discounts, but pricing has more dimensions: model tiers, context length thresholds on some pro models, and media inputs priced differently from text. The Cloud billing console is more powerful than provider dashboards and correspondingly easier to get lost in.
Can Google Cloud billing show me Gemini cost per customer?
No, for the same structural reason as every provider: your customers all call through your credentials, so Google can attribute cost to projects and labels you control, not to users it has never heard of. Per customer cost requires tagging each call with your user id at request time and pricing the token counts yourself, in your own stack.
How do I track Gemini token usage per request?
Every response carries usage metadata with prompt and completion token counts (usageMetadata in the API response). Read it after each call, store it with your user id and feature, and multiply by current per million rates. With the Weckr SDK you wrap the Gemini client once and this happens automatically, including the server side cost recompute at current rates.
Which Gemini model should a cost conscious product use?
The flash lite tier is the budget option (gemini-3.5-flash-lite at about $0.30 input and $2.50 output per million), flash is the mid tier, and pro carries premium rates with context length pricing thresholds. The same routing logic applies as everywhere: route by task difficulty and verify with measured cost per feature rather than assuming.
Keep reading
Three providers, one margin view
Google tells you what Gemini cost your project. Weckr tells you what it cost per customer, next to your OpenAI and Claude spend, against what each customer pays. Two lines per provider, free for 50,000 requests a month. See the cross provider view on the live demo, or start with the AI cost and margin guide.