Gemini for Go Developers - Part 1: The Gemini Model Family · danicat.dev↓<br>Skip to main content<br>danicat.dev
Table of Contents<br>Table of Contents
Welcome to Gemini for Go Developers ! This series is your complete guide to building AI-powered software in Go. Across seven hands-on chapters, we will cover everything from agentic coding, to building autonomous agents with Genkit and ADK , developing games, and using the full G3 Stack (Go, Gemini, GCP) to deploy applications to the cloud.<br>In Chapter 1, we lay the foundation by exploring the Gemini model family, model configurations, and writing our first code with the official Go GenAI SDK.<br>The Gemini model family<br>We often treat “Gemini” as a single name for Google AI offerings, much like using “Google” as a shorthand for search. In reality, Gemini is a family of distinct models built for different operational trade-offs.<br>While frontier models are the ones that capture the headlines, knowing when to use smaller or specialised models is essential for cost-effective engineering. Model selection also directly influences user experience and product adoption, as latency varies sharply across model tiers.<br>Reaching for a Gemini Pro model with high thinking levels for every task is tempting, but it isn’t always the right move. In many cases, it simply increases per-request latency and API costs without delivering a better outcome.<br>Model naming scheme<br>To navigate the Gemini catalog, it helps to understand how Google names its models. A standard model string follows this pattern:
\[<br>\text{[family]}-\text{[version]}-\text{[tier]}{-\text{[modifier]}}<br>\]For example: gemini-3.6-flash or gemini-3-pro-image.<br>Family : While most models will be in the Gemini family, Google also has other model families like Veo and Lyria.<br>Version Numbers : Represent generational leaps in intelligence, context window handling, and instruction adherence.<br>Model Tiers :Pro : Designed for complex multi-step reasoning.<br>Flash : Balanced model with bias towards speed.<br>Flash-Lite : Optimized for speed and high-throughput, simple tasks.
Modifiers : Might indicate a sub-family or specialisation, like image in gemini-3.1-flash-image or live in gemini-3.1-flash-live-preview. It may also include lifecycle modifiers like -preview or -exp (for experimental).<br>Model overview<br>Here is an overview of key Gemini models, starting with the frontier model Gemini 3.x:<br>Gemini 3.x<br>Gemini 3.x is the primary frontier model line, available in Pro, Flash, and Flash-Lite tiers. These general-purpose models are also the primary choice for code generation and software engineering tasks.<br>Current models include:<br>gemini-3.6-flash: High-speed workhorse for multimodal reasoning and agentic tasks<br>gemini-3.5-flash-lite: Lowest-cost, ultra-fast tier for high-throughput microservices<br>gemini-3.1-pro-preview: Advanced tier for complex multi-step reasoning and deep codebase analysis<br>Gemini image models (Nano Banana)<br>While technically still part of the Gemini family, this is a specialised model for image generation, providing both multimodal input and output (image and text). It is capable of producing images from scratch and performing edits on existing images.<br>Current models include:<br>gemini-2.5-flash-image (aka Nano Banana)<br>gemini-3-pro-image (aka Nano Banana Pro)<br>gemini-3.1-flash-image (aka Nano Banana 2)<br>gemini-3.1-flash-lite-image (aka Nano Banana 2 Lite)<br>Veo<br>A model specialised in video generation with native audio. Videos are generated based on text prompts and key images to mark transitions (start and end frame) and as references. Veo 3.1 generates clips up to 8 seconds, but it is possible to extend them up to 20 times in 7-second increments.<br>Current models:<br>veo-3.1-generate-preview<br>veo-3.1-lite-generate-preview (fast generation)<br>Lyria<br>Lyria is specialised in music generation, delivering both instrumental and vocal compositions. Lyria accepts both text and images as input, with the images serving as inspiration for the composition. You can also provide the lyrics yourself or let the model create them for you.<br>Current models:<br>lyria-3-pro-preview<br>lyria-3-clip-preview (short 30s clips)<br>Gemma<br>Gemma is an open-weights model family from Google. It is trained with the same technology behind Gemini, but designed to be deployed on your own infrastructure. Beyond the models offered by Google, Gemma also has a strong community that produces fine-tuned versions for all sorts of use cases.<br>Some Gemma models are small enough to be suitable for running on local machines, enabling use cases where network connectivity is limited or non-existent. The bigger models are very capable, enabling use cases where sovereignty and network isolation are required.<br>Notable mentions<br>Live models: While earlier models handle batch or request-response jobs, Google also offers live models for real-time streaming. Look for -live in the name (e.g. gemini-3.1-flash-live-preview).<br>Text-to-speech: Generates speech from text with narration...