What is AI Studio?

Google AI Studio (aistudio.google.com) is Google’s browser-based tool for experimenting with Gemini models, designing prompts, and generating production-ready code – all before you write a single line yourself. It is the fastest official path from “I want to build something with Gemini” to working code.

Think of it as two things at once: a playground for testing prompts and model behavior interactively, and a launchpad for converting what you build in the browser into real API calls in Python, JavaScript, Go, Java, or other languages. The tool is free to start with no credit card required, which makes it the lowest-friction entry point into Google Cloud AI.

One-line summary: AI Studio is Google’s free, browser-based interface for building and testing prompts with Gemini models, then exporting that work as production API code.

If you need background first: What Is Generative AI? and What Is Artificial Intelligence?

Getting started: what you see first

Go to aistudio.google.com and sign in with a Google account. No billing setup, no API key configuration. The interface opens directly into a chat prompt window.

The left sidebar gives you the main navigation: prompt types (Chat, Realtime streaming, Video gen), saved prompts, and API key management. The main panel is the editor. The right panel – the Run Settings panel – controls which model you are talking to and how it behaves.

The workflow is intentionally simple: write a prompt or system instruction, click Run, see the output, adjust, repeat. When you have something you want to ship, click Get code to export the equivalent API call in your language of choice.

Chat prompts and system instructions

The default mode when you open AI Studio is Chat. This is a multi-turn conversation interface: you type, the model responds, and the full conversation history is included in each subsequent prompt. It maps directly to how you would build a chatbot or any conversational application using the Gemini API.

System instructions

At the top of the chat panel is a field for System Instructions – the persistent instructions that set the model’s behavior before the user conversation begins. This is where you define persona, constraints, tone, output format, and any domain-specific context.

For example, to build a customer support bot that only answers questions about a specific product, you would put the product scope, response format, and escalation instructions here – not in the chat itself. System instructions persist across the conversation and are included in every API call, separate from the user message history.

AI Studio lets you edit system instructions mid-session without starting over, which is useful for iterating on behavior without clearing the conversation.

Few-shot examples

In Chat mode, you can manually insert example exchanges – a user message and the ideal model response – before your actual prompt. This is few-shot prompting: showing the model what good output looks like before asking it to produce its own. AI Studio makes this straightforward by letting you type example turns directly into the conversation timeline.

Model selection and run settings

The Run Settings panel on the right is where you tune how the model behaves. This panel is worth understanding before you build anything serious.

Model selection

A dropdown lets you switch between available Gemini models. As of 2025-2026, the lineup includes Gemini 2.5 Pro (highest capability, best for complex reasoning), Gemini 2.5 Flash (fast and cost-efficient, 1M token context), and Gemini 2.5 Flash-Lite (lowest cost, highest throughput). Switching models mid-session is instant – useful for comparing how different models handle the same prompt.

Temperature

Temperature controls output randomness. At 0, the model is deterministic – the same prompt produces nearly identical output every time. At 1 (or higher for some models), outputs are more varied and creative. For structured tasks like data extraction or code generation, lower is usually better. For brainstorming or creative writing, higher values produce more varied results.

Safety settings

AI Studio exposes four safety filter categories – harassment, hate speech, sexually explicit, dangerous content – each with adjustable thresholds. For most applications you leave these at the default. For specific research or content moderation applications that require seeing borderline content to classify it, the thresholds can be adjusted within the permitted range.

Output length

The Max output tokens slider caps the length of each model response. Setting this appropriately prevents unexpectedly long (and expensive) outputs in production.

Built-in tools: grounding, code execution, and more

Below the model parameters, the Run Settings panel has a Tools section. These are toggles that change what the model can do during a response – each one adds a meaningful capability.

Grounding with Google Search

Toggle on Grounding and the model connects to Google Search before answering. Responses include citations to real web sources, and the model’s answers reflect current information rather than training data alone. This is particularly useful for questions about recent events, current prices, or anything that changes over time.

This is one of AI Studio’s most practical advantages over OpenAI Playground: Search grounding is a single toggle, available for free in the prototyping tier, and built natively into the interface. In OpenAI Playground, web browsing requires different configuration steps and is tied to specific model versions.

Code execution

Toggle on Code execution and the model gains access to a Python sandbox. When the model writes Python code as part of a response, it can actually run that code and return the result – including rendered charts and graphs via Matplotlib. The sandbox supports Numpy, Pandas, and Matplotlib by default. Each code run is limited to 30 seconds; the model can run code up to 5 times per prompt without additional intervention. This makes it useful for data analysis tasks where you want to see actual computed results, not just code suggestions.

Function calling

Toggle on Function calling to define functions (tools) that the model can invoke during a response. You provide the function signatures and descriptions; the model decides when to call them and formats its output as a structured call that your application code can execute. This is the core mechanism for building agents – AI applications that interact with external APIs and services.

In AI Studio, you define functions in a JSON schema directly in the interface and test the model’s calling behavior before writing any application code. This makes it much faster to verify your tool definitions are correct.

Structured output

Toggle on Structured output to constrain the model’s response to a specific JSON schema you define. Instead of prose, the model returns structured data you can parse programmatically. Useful for information extraction, classification tasks, or any application where you need reliable, machine-readable output rather than natural language.

Multimodal: files, images, audio, and video

Gemini models are natively multimodal, and AI Studio exposes this directly. In any prompt, you can attach files alongside text. AI Studio supports images, PDFs (up to 50 MB), audio, video, and other document types (up to 100 MB for general files) via the Files API.

What you can do with this

  • Image analysis: Upload a screenshot, diagram, or photo and ask the model to describe it, extract text from it, or answer questions about what it shows.
  • Document processing: Upload a PDF and ask for a summary, extract specific data points, or have the model answer questions about the document’s content.
  • Audio transcription and analysis: Upload an audio file and ask the model to transcribe it or identify key topics discussed.
  • Video understanding: Upload a short video and ask what happens in it, or at specific timestamps.

Uploaded files persist in the session and can be referenced across multiple prompts. The long context capability of Gemini 2.5 Flash (1 million tokens) means you can pass entire books, long codebases, or hours of transcript into a single prompt and still have room for a detailed response.

Realtime streaming mode

AI Studio also has a Realtime mode, which uses the Gemini Live API. This enables voice-to-voice interaction: speak into your microphone and the model responds with synthesized speech. Useful for prototyping voice assistants or hands-free interfaces. This is separate from the standard Chat mode and uses a different underlying API configuration.

Get code: from prototype to production

The Get code button in the top-right of the editor is what makes AI Studio more than just a playground. Click it and AI Studio generates the exact API call that reproduces your current prompt configuration – system instructions, model settings, tools, file references, and all – in the programming language you choose.

Supported languages: Python, JavaScript, TypeScript, Go, Java, C#, REST (curl), and Google Apps Script.

The generated code imports the correct SDK, sets up the client, constructs the request with your parameters, and handles the response. For a typical chat application this is 10-20 lines of real, runnable code. For more complex configurations with function calling or structured output, the generated code includes the full schema definitions.

This is the intended workflow: prototype the prompt behavior in AI Studio’s browser interface until it does what you want, then export the code and integrate it into your application. You do not have to understand the full Gemini API to start – you explore visually, then switch to code when you are ready.

API key setup

To run the exported code locally, you need an API key. Generate one at aistudio.google.com/app/apikey – it is free and takes under a minute. Set it as an environment variable named GEMINI_API_KEY. The free tier API key works immediately with the rate limits described below.

The new full-stack builder

In March 2026, Google added a significantly more ambitious feature to AI Studio: a full-stack vibe coding experience that generates not just API calls, but entire deployable web applications from a text prompt.

The system is powered by an agent called Antigravity, which builds full applications using Next.js, React, or Angular frameworks. Type a description of what you want to build; the agent scaffolds the frontend, adds backend logic via built-in Firebase integration (Cloud Firestore for databases, Firebase Authentication for logins), installs required libraries automatically, and manages secrets via Google Cloud Secret Manager.

Completed applications can be deployed to Google Cloud Run in one click, or pushed to GitHub for permanent storage. Session state persists across devices, so you can start building on one machine and continue on another.

This positions AI Studio directly against tools like Bolt.new, Lovable, and Replit’s AI features – the “describe the app and get a working product” category. The distinguishing factor is native Firebase and Google Cloud integration, which means authentication and database infrastructure are handled without additional configuration.

Important caveat: this is a newer feature in a category that moves fast. Verify current capabilities at aistudio.google.com before depending on it for production builds.

Free tier: what you actually get

AI Studio’s free tier is genuinely useful for individuals and small teams prototyping applications. No credit card is required to start, and the AI Studio interface itself never incurs charges even after you activate billing for the API.

Model Free RPM Free RPD Free TPM
Gemini 2.5 Pro 5 100 125,000
Gemini 2.5 Flash 10 250 250,000
Gemini 2.5 Flash-Lite 15 1,000 250,000
Gemini 2.0 Flash 15 200 1,000,000

RPM = requests per minute. RPD = requests per day. TPM = tokens per minute. Rate limits apply per project, not per API key. Daily quotas reset at midnight Pacific Time.

One important caveat: on the free tier, your prompts and responses may be used by Google to improve its products. If you are prototyping with sensitive or proprietary data, use a paid API key instead. Activating a paid billing account removes this data usage clause and unlocks higher rate limits.

AI Studio vs. OpenAI Playground

OpenAI Playground (platform.openai.com/playground) is the direct equivalent: a browser-based interface for testing OpenAI models, configuring prompts, and exporting code. Both do the same fundamental job. The differences are real but narrow for most use cases.

Google AI Studio OpenAI Playground
Free tier Yes – no credit card required No – charges at standard API rates
Context window Up to 1M tokens (Flash) Up to 128K tokens (GPT-4o)
Web search grounding Single toggle, free tier included Available, different setup
Prompt versioning Save prompts; no rollback versioning Prompt versioning with rollback
Agent builder Full-stack app builder (Antigravity) Agent canvas for multi-step agents
Free tier data usage May be used to improve Google products Not used to train models by default
Models available Gemini family only GPT family, o-series reasoning models

Choose AI Studio if: you want to start without a billing account, need a 1M-token context window, want Google Search grounding as a simple toggle, or are building on the Google Cloud / Workspace ecosystem.

Choose OpenAI Playground if: you are building on OpenAI models specifically, need prompt version history, or are working with OpenAI’s reasoning models (o-series).

For many developers, the real answer is: use both. Prototyping in AI Studio is free and fast. OpenAI Playground is better if your production app will run on GPT-4o or o-series models. The workflows are similar enough that switching between them is straightforward.

Who it is for (and who it is not)

Good fit

  • Developers starting a Gemini-based project – AI Studio is the fastest path from idea to working API call. Use it to validate that the model does what you need before writing application code.
  • Non-engineers exploring AI capabilities – the interface requires no coding to test prompts. If you want to understand what Gemini can and cannot do before deciding whether to build something, AI Studio is the right starting point.
  • Teams prototyping document or multimodal pipelines – the file upload capability lets you test PDF extraction, image analysis, and audio transcription with real data quickly.
  • Anyone who needs a large context window – 1 million tokens in a free browser tool is genuinely unusual. If you are trying to process very long documents or codebases, this matters.

Not the right tool for

  • Production applications – AI Studio is a prototyping and development tool. Production deployments use the Gemini API directly (or via Vertex AI for enterprise requirements).
  • Teams that need prompt version control – AI Studio lets you save prompts, but it does not have the rollback and comparison versioning that OpenAI Playground offers. If your workflow requires tracking prompt changes over time, account for this.
  • Anything with sensitive or proprietary data on the free tier – the free tier data usage policy means Google may process your inputs. Use a paid API key for anything confidential.

Frequently asked questions

What is Google AI Studio?

Google AI Studio (aistudio.google.com) is Google’s free, browser-based tool for testing Gemini models, designing prompts, and exporting production-ready API code. It lets you experiment with system instructions, model parameters, multimodal inputs, and advanced features like code execution and Google Search grounding – without writing any code upfront. When you are ready to build, clicking “Get code” generates the equivalent Gemini API call in your language of choice.

Is Google AI Studio free?

Yes. AI Studio itself is free to use with no credit card required. The free tier includes access to Gemini 2.5 Flash (10 requests/minute, 250 requests/day) and other models with similar rate limits. The interface never incurs charges even after you activate billing for the Gemini API. The main limitation of the free tier is that your prompts may be used by Google to improve its products – for sensitive data, use a paid API key instead.

How is AI Studio different from Gemini (the chatbot)?

Gemini (gemini.google.com) is a consumer chatbot – you have conversations with it the way you would with ChatGPT. AI Studio is a developer tool – you configure model behavior, test prompts with precise control over parameters, export API code, and build applications. AI Studio exposes settings and capabilities that the Gemini chatbot abstracts away, like temperature, system instructions, function calling, and structured output schemas.

Can I use AI Studio without knowing how to code?

Yes, for prototyping and testing. The chat and prompt interfaces work entirely without code – you type, the model responds, and you adjust settings with sliders and toggles. The full-stack app builder (the Antigravity vibe coding feature) also lets you describe an application in plain English and get a deployable product. You only need to write code when integrating the Gemini API into your own application, which is where the “Get code” export feature becomes relevant.

What is the difference between AI Studio and Vertex AI?

AI Studio is the lightweight developer interface for prototyping with Gemini models quickly – no Google Cloud account required, free to start. Vertex AI is Google’s full enterprise ML platform: it hosts Gemini and other models with enterprise-grade security (VPC-SC, data residency, audit logs), supports custom model training, and provides MLOps infrastructure. For individual developers or small teams building Gemini-powered apps, AI Studio is the right entry point. For enterprises with compliance requirements or teams doing custom ML work, Vertex AI is the correct environment.

Does AI Studio support models other than Gemini?

No. AI Studio is specifically for Gemini models (the various Flash, Flash-Lite, and Pro variants). For access to other models – Anthropic’s Claude, Meta’s Llama, Mistral, and others – you need Vertex AI’s Model Garden, which is a separate product with different access requirements and pricing.

Sources and further reading