Gemini API Free Tier (April 2026): Every Limit, Quota, and Gotcha

Two Ways to Access Gemini for Free

Google offers Gemini through two platforms, each with different free tier mechanics. Understanding the distinction is important because they serve different use cases, and mixing them up leads to unexpected bills or unnecessary limitations.

Google AI Studio (aistudio.google.com)

This is the primary free tier path. You get an API key with zero configuration. No Google Cloud project. No billing account. It works immediately after signing in with a Google account.

AI Studio gives you access to every current Gemini model: 2.5 Flash, 2.5 Pro (experimental), 2.0 Flash, and 1.5 Pro. Each model has different rate limits on the free tier, and we cover those in detail below. The key advantage here is simplicity. You get a key, you make API calls, and you pay nothing until you choose to upgrade.

The tradeoff: Google may use your free-tier inputs and outputs to improve their models. If you are building something with sensitive data, this matters.

Vertex AI ($300 Credit for New Accounts)

Vertex AI is Google Cloud's full ML platform. New GCP accounts get $300 in free credits valid for 90 days. That covers Gemini API calls, but also any other Google Cloud service you use during that window.

After the credits expire, you pay standard rates. There is no ongoing free tier for Vertex AI. The advantage over AI Studio is higher rate limits, no data-sharing clause, enterprise features (grounding, custom fine-tuning, data residency), and SLAs. For most developers starting out, AI Studio is the right choice. Vertex AI makes sense when you need production guarantees or are already in the Google Cloud ecosystem.

Free Tier Rate Limits by Model

These are the limits that actually matter day-to-day. Every model has three constraints: requests per minute (RPM), tokens per minute (TPM), and requests per day (RPD). You will hit the RPM limit long before TPM in most use cases.

A few things stand out. Gemini 2.5 Pro is capped at 50 requests per day on the free tier. That is enough to test and prototype but not enough for any real workload. Gemini 2.0 Flash is the most generous: 15 RPM and 1,500 RPD means you could run a small production app entirely on the free tier if your traffic stays under about 1 request per minute on average.

What the Free Tier Includes Beyond Text

Gemini is multimodal, and the free tier includes all modalities. This is a significant differentiator from OpenAI and Anthropic, where vision capabilities are often limited or cost extra.

• Image understanding: Send images alongside text prompts. Useful for OCR, chart reading, visual Q&A, and image classification. No separate image API needed.
• Video understanding: Upload video clips up to 2 hours in length (with Gemini 1.5 Pro or newer). The model can answer questions about video content, extract information from frames, and summarize visual sequences.
• Audio processing: Send audio files for transcription, translation, and content understanding. Supports common formats including MP3, WAV, and FLAC.
• Document analysis: Upload PDFs directly. The model processes text and images within the document without needing a separate parsing step.
• Code execution: Gemini 2.0 Flash and 2.5 models can execute Python code in a sandboxed environment and return results. This runs inside Google's infrastructure at no additional cost on the free tier.

The rate limits above apply across all modalities. A multimodal request (text + image) counts as one request toward your RPM and RPD limits. Token counts for images and video are calculated based on resolution and duration, which can consume your TPM faster than text-only requests.

Gemini App vs. Gemini API: Different Products

This is where people get confused. The Gemini app (gemini.google.com) and the Gemini API are separate products with separate free tiers.

Gemini app (free): Unlimited text conversations with Gemini. Includes image generation (via Imagen 3). No API access. Think of this as Google's equivalent to free ChatGPT. It uses Gemini 2.0 Flash by default.

Gemini Advanced ($19.99/month via Google One AI Premium): Access to Gemini 2.5 Pro in the app. 2TB Google storage. Gemini in Gmail, Docs, Sheets, and other Workspace apps. NotebookLM Plus. Still no API access included.

Gemini API (Google AI Studio free tier): What this guide covers. Programmatic access to all Gemini models. Completely separate from the Gemini app subscription. You can use both the free app and the free API simultaneously.

A common mistake: people pay for Gemini Advanced thinking it gives them API access. It does not. The API free tier through Google AI Studio is separate and does not require any subscription.

Free Tier vs. Pay-as-You-Go Pricing

When you do outgrow the free tier, here is what you will pay. These rates apply through Google AI Studio with a billing account enabled.

Gemini 2.0 Flash at $0.075 per million input tokens is the cheapest major LLM API on the market. For comparison, GPT-4o mini costs $0.15/$0.60 (twice the input price) and Claude Haiku 4.5 costs $1/$5 (over 13x the input price). If cost is your primary concern and you need a capable model, Gemini 2.0 Flash is difficult to beat even after you leave the free tier.

The jump from free to paid also increases your RPM by 100-200x. That alone justifies upgrading for any production workload.

Practical Tips for Maximizing the Free Tier

If you want to stay on the free tier as long as possible, here are strategies that work.

1. Use Gemini 2.0 Flash for most tasks. It has the highest rate limits (15 RPM, 1M TPM) and is fast. Reserve 2.5 Flash or 2.5 Pro for tasks that genuinely need better reasoning.
2. Cache results aggressively. If your app asks the same types of questions repeatedly, cache the API responses locally. A simple key-value cache (Redis, SQLite, even a dictionary) can eliminate 30-60% of redundant calls.
3. Batch your requests within rate windows. Instead of making calls as they come in, queue requests and send them in controlled bursts that stay under 15 RPM.
4. Use the Gemini app for manual testing. The web app has no rate limits for conversational use. Use it for prompt development and iteration, then switch to the API for production calls.
5. Implement exponential backoff. When you hit rate limits, the API returns 429 errors. A simple retry with exponential backoff (1s, 2s, 4s, 8s) handles transient limit hits gracefully without losing requests.

How Gemini Free Tier Compares to Other Providers

Here is an honest comparison of what you get for free across the major AI API providers as of April 2026.

Google's position is clear: they want developers building on Gemini, and they are willing to subsidize the onboarding. The combination of no expiration, no credit card, multimodal support, and access to competitive models makes the Gemini free tier the strongest starting point for developers in 2026.

For a detailed breakdown of all providers, see our complete AI API free tier comparison.

Common Mistakes to Avoid

After tracking developer experience with Gemini's free tier, these are the mistakes we see most often.

• Confusing AI Studio with Vertex AI. They have different URLs, different authentication, different rate limits, and different billing. Pick one and stick with it.
• Not handling 429 errors. The free tier will rate-limit you. If your code does not have retry logic, requests will silently fail during traffic spikes.
• Sending unnecessarily large prompts. Long system prompts eat into your TPM budget. Keep prompts concise. Use few-shot examples only when they measurably improve output quality.
• Ignoring the data-sharing clause. If you send customer data, PII, or proprietary information through the free tier, Google can use it for training. For production apps with real user data, upgrade to paid or use Vertex AI.
• Using 2.5 Pro when 2.0 Flash would work. The Pro model has a 50 RPD limit on free tier. Most tasks that developers throw at Pro can be handled by Flash with a better prompt. Test with Flash first.

Related Resources