Google AI Studio for New Zealand businesses.

What Google AI Studio is for

Google AI Studio is a browser workspace for trying Gemini models on your own instructions and files. You write a system instruction, adjust settings, run test inputs, and when the results are right, the Get code button turns the setup into Gemini API code. The Run settings panel covers model parameters, safety settings and tools such as structured output, function calling, code execution and grounding with Google Search.

What it does well for a New Zealand business:

• Proving a prompt before anyone writes software. An operations lead can test whether Gemini reads your supplier invoices or job sheets correctly in an afternoon.
• Long documents. Google’s AI Studio page promotes a one-million-token context window, so a full tender pack or policy manual fits in one test. Whether that beats retrieval for your use is covered in RAG versus long context.
• Seeing real costs early. You test on the same models you would pay for, so the numbers carry into a business case.

What it is not: an assistant for staff. It is built for testing prompts and exporting code, and the free tier’s data terms rule out client work. For staff, use Gemini in Google Workspace.

Free tier or paid: what happens to your prompts

This is the decision that matters most, and Google’s Gemini API terms are explicit.

Gemini API terms as published (effective 23 March 2026), checked 19 October 2026.

Free users in the European Economic Area, Switzerland and the UK get the paid-service data terms anyway. New Zealand users do not. Grounding with Google Search adds its own rule: prompts and output are stored for 30 days on either tier.

Under the Privacy Act 2020, the free tier is the clearer problem. Google uses what you submit for its own purposes, so personal information pasted in is disclosed to an offshore provider, engaging IPPs 11 and 12. On a paid project Google acts as your processor, which section 11 treats as you holding the data rather than a disclosure. You still carry the security duty under IPP 5, and processing still happens overseas: Google’s locations list shows Sydney as the nearest Asia Pacific endpoint for its Agent Platform, and says endpoints do not guarantee in-region processing.

What the Gemini API costs

AI Studio itself has no seat price; you pay for API usage on a paid project. Paying requires linking a billing account and prepaying at least US$5 in credits, and API keys stop working when the prepaid balance reaches zero.

Gemini Developer API prices as listed on 19 October 2026 (page updated 16 October 2026). All prices are USD; Google’s page says nothing about New Zealand GST.

Batch requests cost 50% less, with a 24-hour turnaround target. Grounding with Google Search gives 5,000 free requests a month across Gemini 3 and newer models, then US$14 per 1,000.

Illustrative example: 2,000 supplier invoices a month, assuming about 3,000 input tokens and 300 output tokens each. On Gemini 3.8 Flash that is 6 million × US$0.75 plus 0.6 million × US$3.75, or US$4.50 + US$2.25 = US$6.75 a month, about US$3.38 through Batch. From 1 January 2027 the same volume costs US$13.50. On Gemini 4 Argon it is US$18.00, rising to US$36.00 from 1 December 2026. Token fees are rarely the big number; building, checking and maintaining the workflow is. Our guide to choosing between an API and a subscription works through that trade-off.

Starter project: an invoice extraction prototype

Illustrative example: a Hamilton building supplies business keys supplier invoices into its accounts by hand and wants to know whether Gemini can read them reliably.

1. Set up a billed project. Link a Cloud Billing account and prepay the US$5 minimum, so your tests fall under the paid-service terms. If you skip this, use only synthetic invoices.
2. Build a test set of 20 invoices. Choose company-to-company supplier invoices (no sole traders’ personal details) and type the correct answers into a spreadsheet before you start. That spreadsheet is your measure.
3. Open AI Studio and choose Gemini 3.8 Flash. Paste the system instruction below into System Instructions.
4. Turn on structured output in Run settings and enter the schema below, so every answer comes back in the same shape.
5. Run all 20 and score them. Count wrong or missing fields against your spreadsheet. A field that is wrong 1 time in 20 is a review rule, not a rounding error.
6. Repeat on Gemini 3.5 Flash-Lite. If accuracy holds, it costs less per invoice.
7. Click Get code. Hand the code and your scored test set to whoever builds the integration, along with our notes on validating structured outputs.

You extract data from New Zealand supplier invoices.
Return only the fields in the schema. Use null when a field is missing;
never guess. Write dates as DD/MM/YYYY. Write amounts as plain numbers
in NZD. gst_number is the supplier's GST number exactly as printed.
Set check_needed to true if the GST amount is not 15% of the
GST-exclusive total, or if any figure is hard to read.

{
  "type": "object",
  "properties": {
    "supplier_name": {"type": "string"},
    "gst_number": {"type": ["string", "null"]},
    "invoice_number": {"type": "string"},
    "invoice_date": {"type": "string", "description": "DD/MM/YYYY"},
    "total_excl_gst": {"type": ["number", "null"]},
    "gst_amount": {"type": ["number", "null"]},
    "total_incl_gst": {"type": "number"},
    "check_needed": {"type": "boolean"}
  },
  "required": ["supplier_name", "invoice_number", "invoice_date", "total_incl_gst", "check_needed"]
}

Google AI Studio versus the alternatives

• Choose AI Studio to prototype and price a Gemini-based workflow, especially if a developer will build it afterwards.
• Choose Gemini in Workspace for staff who need an assistant inside Gmail, Docs and Sheets with Workspace data protections. Our Google Workspace integration guide covers connecting the two worlds.
• Choose Gemini Enterprise Agent Platform (formerly Vertex AI) for production on Google Cloud with regional endpoints. Gemini 4 Argon there costs US$2/US$10 per million tokens on global endpoints and US$2.20/US$11 on regional endpoints until 30 November 2026.
• Test other vendors too. The same invoice set run through other providers’ consoles gives a fair comparison. If you would rather hand the build over, our AI integration services start from a scored test set like this one.

Where Google AI Studio goes wrong

• Free-tier leakage. Someone pastes a customer complaint into a free account. Make the billed project the only one staff use for work.
• Price cliffs. Argon doubles on 1 December 2026 and 3.8 Flash on 1 January 2027. Budgets built on October prices will be wrong within weeks.
• Model churn. Google deprecated Gemini 3.7 Flash and 3.5 Flash on 8 October 2026 and limits 2.5 models to existing users. Gemini 3.1 Pro is still a preview. Pin a model version and retest when it changes.
• Prepaid balance hits zero. Keys stop working and anything built on them stops too. Set up top-ups and alerts before going live.
• Limits you did not expect. Rate limits apply per project, daily quotas reset at midnight Pacific time (evening in New Zealand), and a new paid project starts at Tier 1, which Google’s rate-limits page lists with a $250 billing tier cap.
• Twenty good results are not proof. A prompt that handles your test set can still fail on the 500th odd invoice. Keep human review on low-confidence results; our lesson on building a no-code AI prototype covers when a prototype is ready to go further.

Written by the Levy Street team in Wellington. Last updated 19 October 2026.