Google's most advanced AI model to date is Gemini. Text, code, graphics, music, and video are just a few of the many forms of information that Gemini can easily understand, process, and integrate because it was designed from the ground up to be multimodal.
MULTIPLE GEMINI SIZES FOR UNMATCHED VERSATILITY
Pro
Description:
The Pro model is the best-performing model designed for a wide range of reasoning tasks. It is versatile, balancing capability and efficiency with a long context window.
Availability:
Available now
Model Versions:
1.0 Pro
1.5 Pro
Strengths:
Versatile, balances capability and efficiency, long context window.
Best Suited For:
General performance across a variety of tasks.
Example Use Cases:
Code generation and improvements
In-depth document analysis
Fine-grained object detection
Long-video summarization and temporal reasoning
Drafting and ideation
Flash
Description:
The Flash model is lightweight and optimized for speed and efficiency. It features low latency, cost-efficiency, and a long context window.
Availability:
Available now.
Model Versions:
1.5 Flash
Strengths:
Lightweight, low latency, cost-efficient, long context window.
Best Suited For:
High-volume, high-frequency tasks at scale.
Example Use Cases:
Summarization
Chat applications
Image and video captioning
Data extraction from long documents and tables
Education products
On-demand content generation (e.g., gaming)
Ultra
Description:
The Ultra model is the largest model, designed for large-scale, highly complex tasks. It offers state-of-the-art performance with advanced reasoning and creativity.
Availability:
Limited - Preview Access with allowlist (Vertex).
Model Versions:
1.0 Ultra
Strengths:
State-of-the-art performance, advanced reasoning and creativity.
Best Suited For:
Research, highly complex tasks, multilinguality.
Nano
Description:
The Nano model is the most efficient model, built for on-device tasks. It emphasizes efficiency, local processing, and offline availability.
Availability:
Limited - Preview Access with allowlist (AI Core).
Model Versions:
1.0 Nano
Strengths:
Efficiency, local processing, offline availability.
Best Suited For:
Research, highly complex tasks, multilinguality.
BUILDING & DEPLOYING GEMINI RESPONSIBLY
From the beginning, we've built Gemini responsibly, integrating security measures and collaborating with partners to enhance its safety and inclusion.
INTEGRATE GEMINI MODELS INTO YOUR APPS WITH:
Google AI Studio
The fastest way to create, experiment, build and launch model prompts with Google's most powerful AI models.
With Google AI Studio, developers can easily integrate Gemini models to quickly create prompts and turn ideas into code to create generative AI applications. A free web-based product called Google AI Studio acts as a gateway to the Gemini API.
You should use AI Studio if you:
Are interested in a free tier to test out and experiment with Gemini
Are looking to prototype and quickly launch apps via an API
Are an independent developer or a developer at an early stage startup
You’re still considering your infrastructure options
Don’t need enterprise-level assistance
Are in a supported region (details)
Vertex AI
For businesses looking for a scalable, fully managed AI platform
Google Cloud's enterprise AI platform, Vertex AI, helps users discover, modify, extend, deploy and manage agents created with the Gemini API. Vertex AI uses the data governance and privacy protections built into Google Cloud, and it also offers tools to help developers use models responsibly and securely.
You should use Vertex AI in Google Cloud if you:
Are an existing Google Cloud customer
Require scale in production
Are an enterprise developer or a developer at a growth stage startup
Need tooling to customize, augment, deploy or govern models
Require enterprise-level support and compliance
Are in a supported region (details)
AICore
A fresh system service that makes Gemini Nano accessible.
A new system-level feature called Android AICore gives users access to Gemini Nano, the smallest version of Gemini, for on-device execution. You can easily create AI experiments with AICore that use sensitive data that are best kept on-device or that work offline. The Google AI Edge SDK includes the APIs required to run Gemini Nano on AICore.
Currently, Google Pixel 8 Pro and Samsung S24 Series are the only devices that support AICore.
You should use Android AICore if you:
Desire to build on-device use cases that benefit from security, limited latency, and ambient experiences
Need to keep sensitive data on-device
You need offline functionality for your use cases
Note: Only Google Pixel 8 Pro and Samsung S24 Series devices are compatible with AI Core.
Integrate Gemini Models Into Your Apps With:
Free Tier: Available with restrictions. Input/output data in the free tier is used to improve Google products.
Rate Limits: Varies by model (See details). Increased rate limits available for paid tier.
Pricing: Varies by model (Gemini API via AI Studio), (Gemini via Vertex).
SDKs Supported: Python, Node.js, Android (Kotlin/Java), Swift, Go for AI Studio; Python, Node.js, Java, Go for Vertex AI.
Develop With Gemini Assistance
Model Capabilities Overview
1.0 Pro
The 1.0 Pro model is designed primarily for text-only tasks. It accepts text inputs and outputs text, making it ideal for a wide range of text-focused applications. This model is the best performing for a variety of text-only tasks.
Context Window:
Vertex: Up to 32K tokens
API via AI Studio:
Free tier: Up to 32K tokens
Paid tier: Up to 120K tokens
Features:
Tuning: Available via Vertex and API via AI Studio
Function Calling: Available via Vertex and API via AI Studio
Context Caching: Not available
JSON Mode: Available on Vertex
JSON Schema: Available on Vertex
System Instructions: Vertex only (See details)
Code Execution: Not available
Vertex Grounding: Available for text input only
1.5 Pro
The 1.5 Pro model supports a wider range of inputs, including text, code, images, audio, video, video with audio, and PDFs. It outputs text and is suitable for tasks that require text or chat prompts to generate text or code responses. This model also supports long-context understanding up to the maximum input token limit.
Context Window:
Vertex: Up to 2 million tokens
API via AI Studio:
Free tier: Up to 32K tokens
Paid tier: Up to 2 million tokens
Features:
Tuning: Coming soon via Vertex and API via AI Studio
Function Calling: Available via Vertex and API via AI Studio
Context Caching: Available via Vertex (coming soon) and API via AI Studio
JSON Mode: Available via Vertex and API via AI Studio
JSON Schema: Available via Vertex and API via AI Studio
System Instructions: Vertex (See details), API via AI Studio
Code Execution: Available via API via AI Studio
Vertex Grounding: Available for text input only
1.5 Flash
The 1.5 Flash model is designed to provide speed and efficiency for high-volume, quality, and cost-effective applications. It supports the same range of inputs as the 1.5 Pro model, including text, code, images, audio, video, video with audio, and PDFs, and it outputs text.
Context Window:
Vertex: Up to 1 million tokens
API via AI Studio:
Free tier: Up to 1 million tokens
Paid tier: Up to 2 million tokens
Features:
Tuning: Coming soon via Vertex and API via AI Studio
Function Calling: Available via Vertex and API via AI Studio
Context Caching: Available via API via AI Studio only
JSON Mode: Available via Vertex and API via AI Studio
JSON Schema: Available via Vertex only
System Instructions: Yes Vertex (See details), API via AI Studio
Code Execution: Yes via API via AI Studio
Vertex Grounding: Available for text input only
FREQUENTLY ASKED QUESTIONS
1. HOW SHOULD I DECIDE WHEN TO USE GEMINI 1.5 PRO VERSUS GEMINI 1.5 FLASH?
Gemini 1.5 Flash is the best option for developers looking to build apps using Google's fastest and most affordable model without sacrificing core quality. For cost-effective deployment of latency-sensitive applications, we recommend Flash. Applications that require the orchestration of complex tasks involving higher-order reasoning with greater depth, knowledge, and granularity than Flash should use Gemini 1.5 Pro. For applications requiring the highest quality for a variety of challenging tasks, we recommend the Pro.
2. WHAT DOES PUBLIC PREVIEW MEAN?
Products or features are developed for customer testing on preview. Although preview products are often made public, they may not be fully functional and may not come with service level agreements or technical support guarantees. Unless otherwise specified by Google, Preview Products are intended to be used exclusively in a test environment. The preview phase usually lasts six months. Learn more: Google Cloud Product Launch Phases and API Versions
03. I WANT TO USE THE FREE TIER FOR THE GEMINI API via GOOGLE AI STUDIO. HOW CAN I ENSURE I AM NOT BILLED?
You can continue to use the Gemini API at no charge if you use it from a Google Cloud project where charging has been removed before May 2, 2024. However, you will not be able to use the features of Google's subscription plan. You may want to create a separate project with billing disabled to use the Gemini API, as disabling billing from an existing project may prevent other billable services. Note: Google's AI Studio still provides free direct access to Gemini models.
4. ARE THERE ANY RESOURCES TO HELP ME START USING THE GEMINI API via GOOGLE AI STUDIO?