Google AI Studio
Last reviewed
May 1, 2026
Sources
25 citations
Review status
Source-backed
Revision
v1 ยท 4,625 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 1, 2026
Sources
25 citations
Review status
Source-backed
Revision
v1 ยท 4,625 words
Add missing citations, update stale details, or suggest a clearer explanation.
Google AI Studio is Google's free, web-based developer interface for prompting, evaluating, and integrating its Gemini and Gemma families of generative AI models. Hosted at aistudio.google.com, the platform serves as the entry-point for individual developers and small teams who want to experiment with the Gemini API before they ship production code or migrate to the enterprise-grade Vertex AI platform on Google Cloud. AI Studio combines a prompt playground, a multimodal file uploader, a code-export panel, and (since late 2025) a full "Build" mode that lets users generate complete React or Node.js applications from natural-language descriptions.
The service was originally launched as MakerSuite at Google I/O on May 10, 2023, when it offered access to the PaLM 2 family. It was rebranded to Google AI Studio on December 13, 2023, the same day Google made the Gemini API publicly available with the launch of Gemini 1.0 Pro. Since then it has grown into the canonical front door for the Gemini Developer API, picking up new models and features at roughly the same cadence as Google's flagship announcements.
AI Studio sits between two other products that share the Gemini brand. The consumer Gemini app (gemini.google.com) is a chat interface aimed at end users, while Vertex AI is the production-grade machine learning platform inside Google Cloud, with VPC controls, IAM, audit logging, and full MLOps tooling. AI Studio is the developer-facing layer in the middle. It is free to start, requires no Google Cloud project, and gives any signed-in Google account an API key within seconds.
The product is positioned as a prototyping and learning tool. A developer can paste a prompt into the playground, attach a PDF or short video, switch between Gemini 3 Pro and Gemini 2.5 Flash with a dropdown, then click "Get code" to copy a Python, JavaScript, Go, or cURL snippet that reproduces the exact request against the public Gemini API. The same API key works in the playground and from a developer's local machine.
| Quick facts | Detail |
|---|---|
| Original name | MakerSuite |
| MakerSuite launch | May 10, 2023 (Google I/O) |
| Rebrand to AI Studio | December 13, 2023 |
| Operator | Google LLC |
| URL | aistudio.google.com |
| Primary models | Gemini 3 Pro, Gemini 2.5 Pro / Flash / Flash-Lite, Imagen 4, Veo 3, Lyria, Gemma |
| Free tier | Yes, with daily and per-minute rate limits |
| Companion API | Gemini Developer API (ai.google.dev) |
| Enterprise alternative | Vertex AI |
| Initial regions blocked | EEA and United Kingdom (later added) |
| Companion SDK | Google Gen AI SDK (Python, JavaScript, Go) |
Google's developer-facing generative AI tooling existed in pieces before AI Studio, scattered across the PaLM API, the Generative AI App Builder inside Google Cloud, and various closed previews. MakerSuite was the first attempt to centralize it. Its successor, AI Studio, has since absorbed nearly every consumer-visible Gemini feature within weeks of its general announcement.
In March 2023, Google opened a limited waitlist for the PaLM API, giving researchers programmatic access to PaLM 2 through a REST endpoint. There was no playground at this stage; developers had to construct JSON payloads by hand. At Google I/O on May 10, 2023, Google introduced MakerSuite as a free browser-based interface for that API. It supported three workflow modes that survived the rebrand: freeform prompts, structured prompts (with input/output examples), and chat prompts. Initial availability was U.S.-only and required joining a waitlist. Through July and August 2023 Google expanded access to roughly 180 countries and added a code-export feature for Python and Node.js.
On December 6, 2023, Google announced Gemini 1.0 in three sizes: Ultra, Pro, and Nano. Seven days later, on December 13, 2023, MakerSuite was rebranded to Google AI Studio and the Gemini API went public. The first publicly available model was gemini-pro (the text variant of Gemini 1.0 Pro), accompanied by gemini-pro-vision for image input and embedding-001 for text embeddings. TechCrunch described the shift as "quite a bit more substantial" than a name change, citing the new vision model, the unified GenerateContent endpoint with streaming responses, and beta features for function calling and semantic retrieval. The initial free tier allowed up to 60 requests per minute. AI Studio was unavailable in the European Economic Area and the United Kingdom at launch, citing GDPR and the EU AI Act; EEA access was added in stages through 2024.
On February 15, 2024, Google announced Gemini 1.5 Pro and made it available in private preview through AI Studio with a context window of up to 1 million tokens. At the time, the largest publicly available context window on any other large language model was 200,000 tokens, so the jump was substantial. AI Studio became the first place outside Google's research teams where developers could feed an entire book, a multi-hour video transcript, or a full code repository into a single prompt.
The rest of 2024 brought a steady stream of additions. Gemini 1.5 Flash entered preview on May 10, 2024, alongside Google I/O announcements. Both 1.5 models reached general availability on May 23, 2024. The 2 million-token context window for 1.5 Pro became GA on June 27, 2024. PDF processing landed in August. Grounding with Google Search was added on October 31, 2024, letting developers cite real-time web results. OpenAI library compatibility came on November 8, 2024. The biggest single jump came on December 11, 2024, when Google introduced Gemini 2.0 Flash Experimental in AI Studio along with the Multimodal Live API, a stateful WebSocket interface for real-time bidirectional voice and video streaming. AI Studio shipped a "Stream Realtime" tab the same day, letting users hold spoken conversations with Gemini while sharing their screen or webcam. A week later, Gemini 2.0 Flash Thinking Mode launched, exposing the model's chain-of-thought reasoning steps.
Gemini 2.0 Flash 001 reached GA on February 5, 2025, with 2.0 Flash-Lite following on February 25. Gemini 2.5 Flash entered preview on April 17, 2025, introducing adaptive thinking budgets that let the model decide how long to reason before answering. Gemini 2.5 Pro and Flash both reached stable status on June 17, 2025. AI Studio also picked up features beyond raw model access: multi-tool use and the URL Context tool on May 20, Batch Mode on July 7, Imagen 4 on August 14, Veo 3 video generation on September 9, the Gemini 2.5 Computer Use Preview (a browser-driving agent) on October 7, and Maps grounding on October 17. Around the same time, Google rolled out Build Mode, an agentic environment that can scaffold a full-stack React/Node.js app from a single prompt, with secrets management and npm support.
On November 18, 2025, Google launched Gemini 3 Pro across the Gemini app, AI Studio, Vertex AI, the Gemini CLI, and Antigravity on the same day. AI Studio served as the primary developer testbed, with Gemini 3 Pro Preview accessible to paid Tier 1 users. Two days later, Nano Banana 2 (the Gemini 3.1 Flash Image Preview) shipped for image generation. The Interactions API entered beta on December 11, 2025, alongside a Deep Research Agent preview. In early 2026 the cadence continued: Gemini 3.1 Pro Preview on February 19, Gemini 3.1 Flash Image Preview on February 26, the first Flash-Lite variant for the 3.x line on March 3, and Lyria 3 music generation on March 25. Google rolled out prepay and postpay billing plans inside AI Studio on March 23, 2026, simplifying upgrades from the free tier. Gemma 4 open-weight models landed on April 2, 2026.
| Date | Event |
|---|---|
| March 2023 | PaLM API limited waitlist opens for PaLM 2 |
| May 10, 2023 | MakerSuite launched at Google I/O |
| December 13, 2023 | MakerSuite rebranded to Google AI Studio; Gemini API public |
| February 15, 2024 | Gemini 1.5 Pro private preview with 1M-token context |
| May 10, 2024 | Gemini 1.5 Flash unveiled at Google I/O |
| May 23, 2024 | Gemini 1.5 Pro and Flash reach GA |
| June 27, 2024 | 2M-token context for 1.5 Pro reaches GA |
| August 9, 2024 | PDF processing added |
| October 31, 2024 | Grounding with Google Search added |
| November 8, 2024 | OpenAI library compatibility introduced |
| December 11, 2024 | Gemini 2.0 Flash Experimental and Multimodal Live API; Stream Realtime |
| December 19, 2024 | Gemini 2.0 Flash Thinking Mode |
| February 5, 2025 | Gemini 2.0 Flash 001 GA |
| April 17, 2025 | Gemini 2.5 Flash Preview with adaptive thinking |
| May 20, 2025 | Multi-tool use; URL Context tool; native audio Live API |
| June 17, 2025 | Gemini 2.5 Pro and Flash reach stable |
| July 7, 2025 | Batch Mode API |
| August 14, 2025 | Imagen 4 GA |
| September 9, 2025 | Veo 3 GA |
| October 7, 2025 | Gemini 2.5 Computer Use Preview |
| October 2025 | Build Mode rolls out ("vibe coding") |
| October 17, 2025 | Grounding with Google Maps GA |
| November 18, 2025 | Gemini 3 Pro Preview launched in AI Studio |
| November 20, 2025 | Nano Banana 2 (Gemini 3.1 Flash Image Preview) released |
| December 11, 2025 | Interactions API beta; Deep Research Agent preview |
| February 19, 2026 | Gemini 3.1 Pro Preview |
| March 3, 2026 | Gemini 3.1 Flash-Lite Preview |
| March 23, 2026 | Prepay and Postpay billing in AI Studio |
| March 25, 2026 | Lyria 3 music generation |
| April 2, 2026 | Gemma 4 models released |
AI Studio is aimed at people who want Gemini API access without the friction of setting up a Google Cloud project. The free tier requires only a Google account, and the API key issued in the AI Studio interface is the same key used by SDKs and cURL clients hitting generativelanguage.googleapis.com. There is no billing setup until the user explicitly chooses to upgrade.
Its closest cousin inside Google's portfolio is Vertex AI Studio, which lives inside the Google Cloud Console and shares many of the same multimodal and tuning features. Vertex AI is the right product when an organization needs region pinning, IAM-based access control, VPC Service Controls, audit logs, model versioning, batch prediction at scale, or data residency guarantees. Google publishes a migration guide for teams that outgrow AI Studio and want to move workloads to Vertex AI without rewriting prompts or SDK calls.
The consumer Gemini app is the third product in the family. It is a chat product for end users, with Gems (custom personas), file uploads for individuals, and a polished mobile experience. AI Studio is for people who plan to call the API from their own software. The same Gemini models power both, but the surfaces are tuned for different audiences. Compared with developer playgrounds from other vendors (the OpenAI Playground, the Anthropic Console, the Hugging Face Inference UI), AI Studio's distinguishing features are the very large context windows, the tight integration with Google Search and Maps grounding, and the unusually rich multimodal input set including video frames, audio waveforms, and screen sharing through the Live API.
AI Studio's interface is organized around a central prompt panel, a model selector on the right, and a left sidebar listing saved prompts, the API key page, and the Build mode. The feature set has expanded considerably since the December 2023 launch; the table below summarizes the current playground.
| Feature | Description |
|---|---|
| Freeform prompt | Single-shot prompting with multimodal input |
| Structured prompt | Tabular input/output examples for few-shot learning |
| Chat prompt | Multi-turn conversation with system instructions |
| Stream Realtime | Bidirectional voice and video using the Live API |
| System instructions | Persistent role, tone, and rule definitions |
| Function calling | Native and OpenAPI-style tool definitions |
| Code Execution tool | Sandboxed Python interpreter Gemini can call mid-response |
| Search grounding | Cite real-time Google Search results in answers |
| Maps grounding | Place data and routing context from Google Maps |
| URL Context | Fetch and reason over arbitrary web pages |
| File Search | Retrieve from a developer-supplied document index |
| Multimodal input | Text, image, audio, video, PDF, and screen capture |
| File uploads | Files API supports up to 2 GB per file, 20 GB per project |
| Long context | Up to 1M tokens on Gemini 1.5 Flash and 2.5 Flash; 2M on 1.5 Pro |
| Thinking mode | Visible chain-of-thought for 2.0 / 2.5 / 3 series |
| Batch Mode | Asynchronous bulk requests at roughly 50% of standard pricing |
| Build Mode | Generate full-stack React/Node.js apps from a prompt |
| Code export | Python, JavaScript / TypeScript, Go, REST cURL |
| Tuning | Limited fine-tuning (PaLM-era; superseded by prompt design and Vertex AI) |
| API key issuance | Single click; no Google Cloud project required |
The Stream Realtime experience is one of the more distinctive demos. Click into it and Gemini opens a microphone connection over WebSockets, listens with voice activity detection, and responds in one of more than 30 voices. With native audio, the model can pick up tone shifts and even tell the difference between a primary speaker and background conversation. It can also receive a video stream from a webcam or a shared browser tab, which is how many of Google's recent product demos have been recorded.
The Build Mode that landed in October 2025 changed AI Studio from a playground into something closer to a lightweight IDE. A developer can describe an app in plain language; the integrated Gemini agent then writes the React frontend and the Node.js backend, manages npm dependencies, stores API secrets, and exposes a preview URL. Internally Google has pitched this as "vibe coding," framing the user's prompt as a description of the desired feel rather than a specification. GitHub integration ships in the same panel, with auto-generated commit messages.
AI Studio is the public surface for almost every Gemini-family model and several Google generative-media models. The exact set rotates as new previews launch and older versions deprecate. The table below reflects the lineup as of April 2026.
| Model | Release | Notes |
|---|---|---|
| Gemini 3.1 Pro Preview | Feb 19, 2026 | Custom tools endpoint; latest reasoning model |
| Gemini 3 Pro | Nov 18, 2025 | Flagship multimodal; "thinking" reasoning |
| Gemini 3.1 Flash Image (Nano Banana 2) | Feb 26, 2026 | High-efficiency image generation |
| Gemini 3.1 Flash-Lite Preview | Mar 3, 2026 | Cost-optimized 3.x variant |
| Gemini 2.5 Pro | Jun 17, 2025 GA | Coding-optimized reasoning model |
| Gemini 2.5 Flash | Jun 17, 2025 GA | General-purpose, adaptive thinking budget |
| Gemini 2.5 Flash-Lite | 2025 | Cheapest 2.5 tier |
| Gemini 2.5 Computer Use | Oct 7, 2025 | Browser-driving agent preview |
| Gemini 2.0 Flash | Feb 5, 2025 GA | Multimodal output; tool use |
| Gemini 2.0 Flash-Lite | Feb 25, 2025 GA | Speed and cost focused |
| Gemini 1.5 Pro | May 23, 2024 GA | Legacy; up to 2M-token context |
| Gemini 1.5 Flash | May 23, 2024 GA | Legacy; up to 1M-token context |
| Imagen 4 (Fast/Standard/Ultra) | Aug 14, 2025 GA | Text-to-image |
| Veo 3 / Veo 3.1 | Sep 9, 2025 GA | Text-to-video; later got extension and reference images |
| Lyria 3 | Mar 25, 2026 | Music generation with text and image input |
| Gemma 1, 2, 3, 4 | Various | Open-weight models also accessible through AI Studio |
| LearnLM | 2024 | Educational reasoning model |
| Gemini Embedding 2 | Apr 22, 2026 GA | Multimodal embeddings (text, image, video, audio, PDF) |
| Gemini Robotics-ER 1.6 | Apr 14, 2026 | Spatial reasoning and instrument reading |
Access to preview models is gated behind the paid tier; free-tier users typically see only the most recent stable Flash and Pro models. The Gemma family is the open-weight cousin of Gemini and can be downloaded for self-hosting in addition to running through the AI Studio interface.
Google has kept the free tier of AI Studio one of the most generous in the industry, but the rate limits have tightened over time. As of early 2026, the free tier provides access to three stable models (Gemini 2.5 Pro, 2.5 Flash, and 2.5 Flash-Lite) at zero cost without requiring a credit card. Limits typically range from 5 to 15 requests per minute and 100 to 1,000 requests per day, with a universal cap of 250,000 tokens per minute across models. In December 2025 Google reduced free-tier limits by roughly 50 to 80 percent, citing capacity pressure following the Gemini 3 launch.
The paid tier (Tier 1) lifts most daily caps, raises per-minute requests to roughly 150-300 depending on the model, and unlocks preview models. Per-token pricing as of April 2026 is summarized below.
| Model | Input ($/M tokens) | Output ($/M tokens) | Notes |
|---|---|---|---|
| Gemini 3.1 Pro Preview | 2.00 / 4.00 | 12.00 / 18.00 | Tiered by prompt length |
| Gemini 2.5 Pro | 1.25 / 2.50 | 10.00 / 15.00 | Tiered by prompt length |
| Gemini 2.5 Flash | 0.30 (text/image/video), 1.00 (audio) | 2.50 | |
| Gemini 2.5 Flash-Lite | 0.10 (text/image/video), 0.30 (audio) | 0.40 | |
| Imagen 4 Fast / Standard / Ultra | n/a | 0.02 / 0.04 / 0.06 per image | |
| Veo 3 Standard | n/a | 0.40 per second of video | |
| Gemini Embedding 2 (text) | 0.20 | n/a | 0.45 image, 6.50 audio, 12.00 video |
Batch Mode requests are billed at roughly 50% of standard rates. Context caching has its own pricing of $0.20-$0.40 per million tokens cached, plus $4.50 per hour of cache storage for Gemini 3.1 Pro. Prepay and postpay billing arrived in AI Studio on March 23, 2026, so individual developers no longer need to wire payment up through Google Cloud Console.
The AI Studio playground sits on top of the Gemini Developer API, which is reachable from any HTTP client. Google publishes first-party SDKs in three languages: google-genai for Python (replacing the legacy google-generativeai), @google/genai for JavaScript/TypeScript (replacing @google/generative-ai), and google.golang.org/genai for Go. A REST endpoint at generativelanguage.googleapis.com is always available. The Gen AI SDK gives the same code path access to either the Gemini Developer API used by AI Studio or the Gemini API on Vertex AI, switching transports based on a configuration flag, so a prototype written against an AI Studio API key can move to a Vertex AI deployment without major rewrites.
Third-party integrations have caught up. LangChain ships both Python and JavaScript adapters for the Gen AI SDK. LlamaIndex, Haystack, and AutoGen all maintain Gemini connectors. The November 2024 OpenAI compatibility layer means any application written against the OpenAI Python or Node SDK can point at https://generativelanguage.googleapis.com/v1beta/openai/ and receive Gemini responses with no code changes beyond model name and base URL.
Every major foundation model lab now ships a developer-facing playground. AI Studio's positioning is most easily understood by comparing it with the alternatives.
| Tool | Vendor | Models | Free tier | Code export | Multimodal | Notable |
|---|---|---|---|---|---|---|
| Google AI Studio | Gemini, Gemma, Imagen, Veo, Lyria | Yes (rate-limited) | Python, JS, Go, cURL | Text, image, audio, video, PDF, screen | Live API; Build Mode | |
| Vertex AI Studio | Same plus Anthropic, Mistral, Meta | No (Cloud billing required) | Python, REST | Same as AI Studio | Enterprise IAM, VPC, MLOps | |
| OpenAI Playground | OpenAI | GPT-4.1, GPT-5, o-series | No (paid only) | Python, Node, cURL | Text, image, voice | Assistants API; tool calling |
| Anthropic Console / Workbench | Anthropic | Claude 4 / 4.5 / 4.7 | Limited credit | Python, TS, cURL | Text, image, PDF | Prompt library; evals |
| Hugging Face Spaces / Inference Endpoints | Hugging Face | Open-source plus partners | Free demos | Python, JS | Varies by model | Model hub integration |
| Azure AI Studio (Foundry) | Microsoft | Azure OpenAI, Llama, Mistral, Phi | No | Python, C#, REST | Text, image, voice | Tied to Azure subscription |
| Amazon Bedrock | AWS | Anthropic, Meta, Mistral, Amazon Nova | Limited | Python, Java, Node | Varies | Multi-vendor; AWS billing |
AI Studio's stronger differentiators are the very large free-tier context window, the unified multimodal interface (image, video, audio, screen), and the depth of Google ecosystem grounding (Search, Maps, eventually YouTube). Its weak spots compared with the enterprise tools are the absence of VPC-style isolation and the fact that data on the free tier may be used to improve Google's products.
Rapid prototyping for Gemini-based apps was the original purpose: a developer pastes a prompt, iterates on system instructions, and exports working code in under a minute. Long-context experiments took over after 1.5 Pro launched, with users uploading entire books, code repositories, or video transcripts and asking the model to summarize, refactor, or answer questions across the full corpus. Multimodal reasoning experiments combine a screenshot, an audio clip, and a written question in a single prompt. Voice-agent prototyping happens through the Live API and Stream Realtime tab. Function-calling design and debugging is faster in the playground than from a local terminal. Many of Google's own product reveal videos, including the original Gemini multimodal demo, were recorded inside AI Studio. Free access and code export make it the default environment for university courses and student hackathons. Teams that begin in AI Studio can later move the same code, with the same SDK, onto Vertex AI for production.
The data terms differ between the free and paid tiers. On the free tier, Google states that prompts, files, and responses may be reviewed by humans (de-identified from accounts and API keys) and used to improve Google's products. On the paid tier, accessed through Google Cloud Billing, AI Studio data is treated under Google's Data Processing Addendum and is not used to train Google's models. This split mirrors the practice of most other commercial AI APIs.
Regional availability has expanded over time. AI Studio launched without EEA or UK access in December 2023 because of GDPR and EU AI Act considerations. EU access was added in stages through 2024 and 2025. Some preview features and models continue to roll out outside the EU first. AI Studio also enforces a minimum age requirement (18 in most jurisdictions) for account holders.
For stricter compliance environments (HIPAA, FedRAMP, customer-managed encryption keys, audit logs, region pinning), the recommended path is to use Vertex AI instead of AI Studio. Google explicitly documents this distinction; AI Studio does not currently expose VPC Service Controls, customer-managed encryption keys, or audit logs.
The pace of feature additions has been steady enough that what felt like a generous prompt sandbox in 2024 looks like a lightweight IDE in 2026. Highlights of the past year and a half include native function calling that can chain multiple tool calls in one response, default-on Google Search and Maps grounding, the Live API's move to native audio with 30+ voices, Build Mode and the Antigravity Agent for full-stack code generation, same-day access to Gemini 3 Pro and 3.1 Pro, an MCP server integration, file search across user-supplied document indexes, and prepay/postpay billing inside AI Studio itself.
AI Studio is intentionally lighter than Vertex AI, and that has trade-offs. There is no private VPC option, no IAM, and no audit logs, so anything that needs network isolation belongs on Vertex AI. Fine-tuning is limited; the PaLM-era tuning UI was scaled back when Gemini launched, and full-scale supervised fine-tuning lives on Vertex AI. Free-tier rate limits can be restrictive, especially after the December 2025 reduction, so production traffic should not depend on them. Batch prediction does not match Vertex AI scale, although AI Studio now has Batch Mode for bulk inference. The interface is browser-based and personal; it lacks team-level governance, role-based access, or shared prompt libraries with audit history. Some preview features and models are restricted to the paid tier or to specific regions. Data on the free tier may be used to improve Google products, so anyone working with sensitive data should upgrade to paid or move to Vertex AI.