Google AI Studio
Google AI Studio is the fastest zero-friction way to get a working Gemini API key and prototype multimodal prompts in a browser, with no credit card. The catch is structural: the free tier excludes Gemini 4 Argon and all image/video models, trains on your inputs with no self-serve opt-out, and hides rate limits inside a console; paid rates for Flash are promotional and double in 2027; thinking tokens are billed as output; and AI Studio has no IAM, VPC-SC, SLAs, or regional endpoints—those live in Vertex AI, which is a separate migration.
What is Google AI Studio?
Google AI Studio is Google's front door to the Gemini API. It is not a subscription product and not a standalone model: it is a free, browser-based developer environment where you write and test prompts against Gemini models, tune parameters (temperature, top-p, top-k, safety settings), feed in text, images, audio, video, and PDFs, and—when you are ready—generate a real API key without leaving the browser or entering a credit card. The same surface includes a Build mode that takes a plain-English description and scaffolds a working front-end wired to the Gemini API, which you can preview and iterate on, and optionally deploy to Cloud Run.
The product's identity is inseparable from Google's two-entry-point strategy for generative AI. Google splits its generative-AI surface into two doors over largely the same Gemini models: AI Studio, which is designed for prototyping, learning, and 'hello world'; and Vertex AI (now Gemini Enterprise Agent Platform), which is designed for running AI as governed enterprise infrastructure on Google Cloud. The official documentation describes the split in exactly those terms. AI Studio has an API key, a free tier, and almost no governance. Vertex has IAM, VPC Service Controls, CMEK, audit logs, SLAs, HIPAA, regional endpoints, and a contractual commitment not to train on your data. They converge through a documented migration path—the same code runs against Vertex with different client credentials—but they are not the same product, and the gap between them is the source of most of AI Studio's limitations.
The free tier is genuinely free, and it is genuinely limited. It covers Gemini 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, Gemma 4, several Live and TTS models, and Gemini Embedding. It does not cover Gemini 4 Argon, Gemini 3.1 Pro Preview, any image model, Veo 3.1, Omni Flash, or Lyria. The models that make Gemini competitive for serious reasoning and agentic work are behind the paid wall. And the free tier's data terms are explicit: free-tier prompts and responses may be used to improve Google products, including model training, with no documented self-serve setting to disable that use. Linking a billing account switches the project to paid tiers and stops the data-improvement use, but it also means every token is metered—there is no way to keep the free allowance and opt out of training.
Paid pricing is per token and changes on fixed dates. Gemini 4 Argon is $2 input and $10 output per 1M tokens through November 30, 2026, then $4/$20 from December 1. Gemini 3.8 Flash is $0.75/$3.75 through December 31, 2026, then $1.50/$7.50 from January 1, 2027. Both output rates include thinking tokens, which are not optional and are not separately disclosed per request. Grounding with Google Search adds $14 per 1,000 requests after a shared 5,000/month allowance, and one user request can trigger multiple search queries, each billed separately. Batch API halves the cost but is paid-tier only. Google Cloud Free Trial credits do not apply to AI Studio; they only work through Vertex AI.
How does Google AI Studio work?
A developer signs in with a Google account, opens aistudio.google.com, and selects a model. The core interface is a prompt panel with system instructions, a model picker, and labeled controls for temperature, top-p, top-k, max output tokens, stop sequences, and safety settings per harm category. Multimodal inputs can be dropped in directly: an image for visual Q&A, a PDF for structured extraction, a short video for timestamped breakdowns. Prompts and configurations can be saved and reused. The single most important button is 'Get API key,' which generates a Gemini API key tied to the Google account in under a minute, with no billing setup required for the free tier.
Build mode sits on top of the playground. A user describes an app in plain language; AI Studio generates working front-end code wired to the Gemini API, previews it in the browser, and allows iterative edits. The generated app can be deployed to Cloud Run from within AI Studio, but the Starter tier caps deployment at two services in a single Cloud Run region, and advanced deployment features require a paid API key. The builder's API key and quota pay for end-user calls in shared and deployed apps, so public abuse can drain the creator's quota.
For developers going past casual prompting, AI Studio exposes function calling (the model can request that your code run a specific action), structured JSON output (responses can be parsed reliably), and grounding with Google Search (answers reflect current web information rather than training data). These are the features that make the jump from demo to integration, and testing them here—before writing a line of integration code—is the point of the tool. When the project needs IAM, VPC-SC, CMEK, audit logs, SLAs, HIPAA, or regional endpoints, the migration path is to Vertex AI / Agent Platform: the same code with `genai.Client(api_key=...)` becoming `genai.Client(vertexai=True, project=..., location=...)`.
Use Cases
- Getting a working Gemini API key in under a minute and testing prompts against Flash, Flash-Lite, Gemma, Live, TTS, and Embedding models without a credit card
- Prototyping multimodal workflows—PDF extraction, image Q&A, video breakdowns—before committing to integration code
- Testing function calling, structured JSON output, and Google Search grounding in the same interface where you tune temperature and safety settings
- Scaffolding a front-end demo with Build mode and iterating on it in the browser, accepting that deployment is limited and the output is a prototype
- Evaluating whether Gemini's Flash-class models are sufficient for a workload before moving to Argon or Vertex AI
- Learning the Gemini API surface with zero financial commitment, then migrating the same code to Vertex AI when governance requirements appear
Pricing
Pricing model: Freemium
| Plan | Price | Billing | Included usage | Notes |
|---|---|---|---|---|
| Free tier | $0 | per month | Free input and output tokens on selected models: Gemini 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, Gemma 4, several Live and TTS models, and Gemini Embedding. Gemini 4 Argon, Gemini 3.1 Pro Preview, image models (Nano Banana), Omni Flash, Veo 3.1, and Lyria are not available on the free tier. Rate limits are no longer published per model; they are shown per project in the AI Studio rate-limit console and are more restricted for experimental and preview models. No credit card required. Free-tier prompts and responses may be used to improve Google products, with no documented self-serve opt-out. | — |
| Paid tier (Gemini 4 Argon) | $2 / $10 | per 1M input / output tokens | Promotional rates through November 30, 2026: $2 input and $10 output per 1M tokens. From December 1, 2026: $4 input and $20 output. Output price includes thinking tokens. Context caching: $0.10 per 1M through Nov 30, $0.20 after. Implicit caching only; explicit caching is not supported. Grounding with Google Search: 5,000 free requests/month shared across Gemini 3+ models, then $14/1,000 requests, capped at 1,500/day. Every token is metered, including thinking tokens and internal tool system instructions when using grounding. | — |
| Paid tier (Gemini 3.8 Flash) | $0.75 / $3.75 | per 1M input / output tokens | Introductory rates through December 31, 2026: $0.75 input and $3.75 output per 1M tokens. From January 1, 2027: $1.50 input and $7.50 output. Context caching $0.075 per 1M through Dec 31, $0.15 after. Grounding with Google Search available on paid tier at 5,000 free requests/month, then $14/1,000. | — |
| Paid tier (Gemini 3.5 Flash-Lite) | $0.30 / $2.50 | per 1M input / output tokens | Input $0.30 per 1M (text/image/video/audio), output $2.50 per 1M including thinking tokens. Context caching $0.03 per 1M. Grounding with Google Search available on paid tier at 5,000 free requests/month. | — |
| Enterprise (via Gemini Enterprise Agent Platform) | Custom | annual | Same models plus third-party and open-source models through Vertex AI / Gemini Enterprise Agent Platform. IAM, VPC Service Controls, CMEK, audit logging, 24/7 support, SLAs, HIPAA, SOC 2, provisioned throughput, volume discounts, regional endpoints, Model Registry, and MLOps tooling. Prompts and responses are not used for training. Contact sales. | — |
Common Limitations
- The free tier cannot access Gemini 4 Argon, and the models it does offer are on promotional clocks · The free tier's model list is explicitly narrow: Gemini 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, Gemma 4, several Live and TTS models, and Gemini Embedding. Gemini 4 Argon—the flagship reasoning model built for complex software engineering, deep math, and long-horizon agents—is marked 'Not available' on the free tier. Gemini 3.1 Pro Preview, all image models, Veo 3.1, Omni Flash, and Lyria are also excluded. This is not a temporary gap: the free tier is positioned as a limited-access playground, and the models that make Gemini competitive for serious agentic work are behind the paid wall. Worse, the paid rates for the models that are free today are promotional. Gemini 3.8 Flash is $0.75/$3.75 until December 31, 2026, then $1.50/$7.50. A team that prototypes on the free tier with Flash and then moves to production will not only need to pay, but may face a doubled rate card within weeks of starting.
- Free-tier data is used for product improvement, and there is no self-serve opt-out · Google's pricing page is unusually blunt: the free tier column for 'Used to improve our products' says Yes, while the paid tier says No. The Gemini API Additional Terms state that when you use unpaid services, Google uses the content you send and the responses generated to provide, improve, and develop Google products, services, and machine learning technologies. Multiple independent sources confirm that the free version of AI Studio does not offer a setting to disable this, and that human reviewers may examine inputs for quality purposes. The practical consequence is that confidential information, customer PII, proprietary code, or internal documents entered into the free tier should be treated as potentially entering Google's training and review pipeline. Linking a billing account changes the data posture to 'No' for the paid tier, but this is a one-way switch: you cannot keep the free tier's token allowance while opting out of data use. For teams handling sensitive material, the free tier is not a compliant starting point.
- Rate limits are unpublished, low, and enforce per project, not per key · Google no longer publishes per-model RPM or RPD figures on its rate-limits page. The active limits are shown only inside the AI Studio rate-limit console for each project, and the documentation states that experimental and preview models are limited more tightly. Third-party measurements in October 2026 placed Gemini 3.8 Flash at approximately 5 requests per minute and 20 requests per day, with some models reaching 5–15 RPM and up to 500 RPD—but these are not official numbers, and they can change without notice. The limits apply per project, not per API key, so creating additional keys within the same project does not multiply the quota. A developer testing a multi-turn agent loop or a batch of prompts can exhaust the daily allowance in minutes without any warning from the console. The absence of published limits also makes capacity planning impossible: you cannot read a table, multiply, and know what your free-tier ceiling is. You have to discover it by hitting a 429.
- Thinking tokens are billed as output, and they are not disclosed per request · Every reasoning model on the paid tier bills thinking tokens as output tokens. The pricing page states this explicitly: 'Output price (including thinking tokens).' For Gemini 4 Argon, that means the $10 per 1M output rate through November 30 (and $20 after) applies to the model's internal reasoning, not just the visible answer. A complex prompt that triggers a long chain of thought can consume far more output tokens than the response text suggests. The API does not expose a simple per-request dollar figure before the request runs, and the model's thinking length is not user-controllable in the way some competitors offer reasoning-effort parameters. A team budgeting from input/output token counts on visible text will systematically underestimate the bill on Argon, especially for agentic or mathematical workloads where the model is designed to think longer. Flash models also bill thinking tokens, though their lower rate makes the absolute impact smaller.
- AI Studio is a prototype surface, not a production platform: no IAM, VPC-SC, CMEK, SLAs, or regional endpoints · AI Studio's authentication is an API key tied to a Google account or OAuth; it has no IAM, no VPC Service Controls, no customer-managed encryption keys, no audit logging, no 24/7 support, and no SLAs. The endpoint is global only, with no strict regional endpoints. Google's own documentation splits the product into two entry points over largely the same Gemini models: AI Studio is 'open the hood and tinker with Gemini for free,' while Vertex AI / Gemini Enterprise Agent Platform is 'run AI as governed enterprise infrastructure on GCP.' The migration path exists—the same code runs against Vertex with different client credentials—but it is a migration, not a toggle. Teams that build a workflow in AI Studio and later need IAM, VPC-SC, HIPAA, or a regional endpoint must move to Vertex AI, which uses a different endpoint (aiplatform.googleapis.com), different authentication, and a different billing structure. Google Cloud Free Trial credits do not apply to AI Studio at all; they only work through Vertex.
- Build mode's deployment is capped at two services and a single region on the Starter tier, and public apps spend the creator's quota · AI Studio's Build mode can generate and deploy a full-stack app to Cloud Run, but the Starter tier imposes hard limits: up to two services, deployed in a single Cloud Run region. Advanced features like Veo 3.1 and Cloud Run deployment require a paid API key. More importantly, the builder's Gemini API key and quota pay for end-user calls in shared and deployed apps—so a public-facing app that goes viral, or is abused, can drain the creator's quota and create availability or cost problems. Independent reviews also note that the generated apps are prototypes: complex features like maps, chat, and live data break easily, security and compliance gaps remain, and AI-generated code still requires human review, tests, dependency checks, and hardening before production use. Build mode is a fast prototyper, not a deployment platform, and the gap between a working demo and a production app is where most of the engineering work still lives.
- Single-model lock-in: everything routes through Gemini · AI Studio is a Gemini playground, not a multi-model evaluation stack. Every prompt you run goes to a Gemini model; there is no interface for comparing Gemini against Claude, GPT, or an open-weight model in the same session. If your workflow depends on model comparison, fallback across providers, or choosing the cheapest model for each task, AI Studio cannot be the only tool in the stack. The API key you generate is a Gemini API key, and the pricing, rate limits, and model availability are all Gemini-specific. This is not a flaw in execution—it is the product's scope—but it means AI Studio is the right first stop for building on Gemini, not a substitute for a gateway or a multi-model evaluation layer.
Key Features
- Free, browser-based prompt playground for Gemini models with no credit card required
- One-click Gemini API key generation without a Google Cloud project for the free tier
- Multimodal inputs: text, images, audio, video, PDFs
- Parameter controls: temperature, top-p, top-k, max output tokens, stop sequences, safety settings per harm category
- Function calling, structured JSON output, and grounding with Google Search
Google AI Studio vs. Vertex AI and OpenRouter
This comparison uses official pricing pages, documentation, and independent reviews available as of November 2026. It focuses on the structural difference that determines which entry point a team should use: governance and model scope. Google AI Studio is a zero-friction, free-tier Gemini playground with no enterprise controls; Vertex AI is the same models behind Google Cloud's governance, SLAs, and regional endpoints; OpenRouter is a multi-provider gateway that solves model breadth rather than Google-specific governance. AI Studio and Vertex are two doors to the same models; OpenRouter is a different building.
| Tool | Positioning | Key differences | Best fit | Pricing note |
|---|---|---|---|---|
| Google AI Studio Official Gemini Developer API pricing Free-tier model availability and data terms (independent review, 2026) AI Studio vs Vertex decision guide (independent, 2026) | Google's free, browser-based prompt playground, API onboarding tool, and lightweight app builder for the Gemini model family, designed for prototyping rather than production governance. | The fastest path from zero to a working Gemini API key, with no credit card and a genuinely usable free tier for Flash, Flash-Lite, Gemma, and audio/embedding models. But the free tier excludes Gemini 4 Argon, all image and video models; trains on inputs with no self-serve opt-out; publishes no rate limits outside the per-project console; bills thinking tokens as output on paid models; and lacks IAM, VPC-SC, CMEK, audit logs, SLAs, and regional endpoints. Those live in Vertex AI. AI Studio is single-model-family: every prompt routes through Gemini. | Developers who want to test Gemini prompts and multimodal inputs with zero setup, get an API key in under a minute, and decide whether Gemini is the right model before committing to Google Cloud infrastructure or a paid contract. | Free tier: $0, selected Flash/Flash-Lite/Gemma/Live/TTS/Embedding models, unpublished rate limits, data used for improvement. Paid: Gemini 4 Argon $2/$10 per 1M tokens through Nov 30, 2026, then $4/$20; Gemini 3.8 Flash $0.75/$3.75 through Dec 31, 2026, then $1.50/$7.50; Flash-Lite $0.30/$2.50. Thinking tokens billed as output. Grounding search $14/1,000 after 5,000/month. Batch API halves cost, paid-tier only. |
| Vertex AI (Gemini Enterprise Agent Platform) Migration guide: AI Studio to Agent Platform (Google Skills, official) AI Studio vs Vertex decision guide (independent, 2026) | Google Cloud's enterprise AI platform, running largely the same Gemini models as AI Studio but with IAM, VPC Service Controls, CMEK, audit logs, SLAs, HIPAA, regional endpoints, and a contractual commitment not to train on customer data. | Vertex is the production half of Google's two-entry-point strategy. It uses a different endpoint (aiplatform.googleapis.com), different authentication (IAM service accounts, OAuth 2.0, VPC-SC), and a different billing structure (Google Cloud projects, provisioned throughput, volume discounts). It supports third-party and open-source models in addition to Gemini, and it offers Model Registry, Model Monitoring, pipeline evaluation, and 24/7 enterprise support. Google Cloud Free Trial credits apply here but not to AI Studio. The migration path from AI Studio is documented and requires code changes only at the client-credential level, but the governance model and cost structure are different products. Vertex is not a free-tier playground; it is infrastructure, and it is priced and governed accordingly. | Enterprises and production teams that need IAM, VPC-SC, CMEK, audit logs, SLAs, HIPAA, SOC 2, regional endpoints, or a contractual no-training commitment, and that are willing to operate inside Google Cloud's governance and billing model. | Same Gemini per-token rates as AI Studio's paid tier, plus Google Cloud billing, provisioned throughput options, and volume discounts. Google Cloud Free Trial credits ($300) apply. Enterprise support and SLAs are contract-based. No free-tier playground equivalent. |
| OpenRouter Official BYOK documentation Official data collection policy | A managed multi-provider LLM gateway that puts 500+ models from 80+ providers behind one OpenAI-compatible API, with automatic fallback and no per-token markup, optimized for model breadth rather than Google-specific governance. | OpenRouter solves the opposite problem from AI Studio. AI Studio is a single-model-family playground with a free tier and no enterprise controls; OpenRouter is a multi-model gateway with no free playground and a credit-based fee structure (5.5% card fee, $0.80 minimum; 8% Business; 5% crypto). OpenRouter does not train on prompts by default and offers EU/US in-region routing on Business, which AI Studio does not. But OpenRouter is not self-hostable and has no IAM/VPC-SC comparable to Vertex. A team that wants Gemini specifically and a zero-friction start should use AI Studio; a team that wants to compare Gemini against Claude, GPT, and open-weight models through one endpoint should use OpenRouter. They are complements, not substitutes. | Teams that need model breadth, automatic fallback across providers, and a single OpenAI-compatible endpoint, and that are comfortable with a managed gateway rather than a single vendor's playground or a cloud provider's governance stack. | Free: 25+ free models, 20 req/min, 50 req/day. Standard: token passthrough, 5.5% card fee ($0.80 min), 5% crypto, BYOK free to $25K/month. Business: 8% fee, EU/US in-region routing. Enterprise: custom, SSO/SCIM, SLAs, ZDR guarantees. |
Comparison checked against official product information on 2026-11-04. Features and pricing may change.
Frequently Asked Questions
Is Google AI Studio really free?
The AI Studio interface, prompt playground, and Build mode are free, and the free tier provides free input and output tokens on selected models without a credit card. But the free tier excludes Gemini 4 Argon, Gemini 3.1 Pro Preview, all image models, all video models (Veo 3.1, Omni Flash), and Lyria. It also has unpublished rate limits that are shown only inside the per-project console, and free-tier prompts and responses may be used to improve Google products, including model training, with no documented self-serve opt-out. It is a genuinely usable prototyping tier, not a free production tier.
Can I use Gemini 4 Argon for free in AI Studio?
No. Gemini 4 Argon is marked 'Not available' on the free tier. It is paid-only: $2 input and $10 output per 1M tokens through November 30, 2026, then $4/$20 from December 1. Every token is billed, including thinking tokens and internal tool system instructions when using grounding. If Argon is the model your workload needs, the free tier cannot evaluate it; you need a linked billing account.
Does Google train on my AI Studio prompts?
On the free tier, yes—Google's pricing page states that free-tier content is used to improve Google products, and the Gemini API Additional Terms say Google uses unpaid-service content to provide, improve, and develop its products and machine learning technologies. There is no documented self-serve setting to disable this on the free tier. Linking a billing account switches the project to the paid tier, where the 'Used to improve our products' column says No. But that is a one-way switch: you cannot keep the free allowance and opt out of data use. For confidential or regulated data, the free tier is not a compliant starting point.
What are the free-tier rate limits?
Google no longer publishes per-model RPM or RPD figures. The active limits are shown only inside the AI Studio rate-limit console for each project, and experimental or preview models are limited more tightly. Third-party measurements in October 2026 placed Gemini 3.8 Flash at roughly 5 RPM and 20 RPD, with some models up to 5–15 RPM and 500 RPD, but Google does not confirm these numbers and they can change. The limits apply per project, not per API key, so adding keys within the same project does not raise the ceiling. The practical experience is discovering the limit by hitting a 429, not by reading a table.
What is the difference between Google AI Studio and Vertex AI?
They are two entry points over largely the same Gemini models, with different governance and data terms. AI Studio is for prototyping: API key auth, free tier, no IAM, no VPC-SC, no CMEK, no audit logs, no SLAs, no regional endpoints, and free-tier data may be used for product improvement. Vertex AI (Gemini Enterprise Agent Platform) is for production: Google Cloud IAM, VPC Service Controls, CMEK, audit logging, 24/7 support, SLAs, HIPAA, SOC 2, regional endpoints, provisioned throughput, and a contractual commitment not to train on prompts or responses. The migration path is documented—same code, different client credentials—but it is a migration, not a toggle. Google Cloud Free Trial credits apply only to Vertex, not AI Studio.
Alternative Tools
-
Composio Composio is a managed tool-integration and authentication layer that exposes 1,500+ apps as agent-callable tools via SDKs, a CLI, or hosted MCP endpoints, handling OAuth, tool search, and execution for AI agents. -
Context7 Context7 is an Upstash-built MCP server that injects up-to-date, version-specific library documentation and code snippets into AI coding agents, reducing deprecated-API hallucinations for indexed frameworks and packages. -
Firecrawl Firecrawl is the most complete LLM-ready web data API: six endpoints covering search, scrape, crawl, map, parse, and interactive browser sessions, with clean Markdown output and first-class MCP support. The advertised 1-credit-per-page rate is a base rate—JSON extraction, PDF parsing, ZDR, and prompt-injection checking multiply it by up to 9x, and Interact bills by the browser minute. Self-hosting is AGPL-3.0 but omits the anti-bot, Agent, and Interact features that define the cloud product. Plan credits do not roll over below $599/month. -
OpenRouter OpenRouter is the widest managed gateway to LLMs: 500+ models across 80+ providers behind one OpenAI-compatible endpoint, with automatic fallback, MCP support, and no per-token markup. Its pricing is honest about the token pass-through but less forthcoming about the $0.80 minimum fee that makes small top-ups cost 8–16% effectively, and about the fact that EU/US in-region routing requires the 8% Business tier. BYOK is free up to $25,000/month, which is a non-issue for almost everyone. It is a strong default for model breadth and zero-ops access—not for self-hosting, small-budget cost optimization, or contractual uptime guarantees. -
Model Context Protocol MCP is the de facto open standard for connecting AI agents to tools and data, adopted by every major AI platform and stewarded by the Linux Foundation. It solves the integration transport problem well. It does not solve the security problem: tool poisoning succeeds at 72.8–84.2% rates against leading models, the specification has no context isolation or least-privilege enforcement, and 49% of enterprises cite security and compliance as their top deployment challenge. -
Qwen Chat Qwen Chat is Alibaba's free consumer AI assistant, offering 1M-token context, multimodal understanding, image and video generation, Deep Research, and web development tools, with optional paid consumer tiers for higher quotas.