AI / Google Antigravity Gemini Fundamentals Interview Questions
1. What is the Gemini API and what does it give developers access to?
The Gemini API is Google's developer interface to its Gemini family of multimodal AI models. It allows developers to integrate state-of-the-art language, vision, audio, image generation, and video capabilities into applications via simple HTTP calls or official SDKs. Access is through Google AI S...
2. What is the Interactions API and how does it differ from generateContent?
The Interactions API is Google's new recommended interface for building with Gemini models and agents, generally available as of June 2026. The original generateContent API remains fully supported but is now considered legacy for new projects. Interactions API vs generateContent Feature generateC...
3. What are the current Gemini model families and which should you use for different tasks?
As of mid-2026, the Gemini model ecosystem spans multiple generations. The Gemini 3.x series is the primary production line, with Gemini 3.5 Flash being the newest generally-available flagship. Current Gemini models mid-2026 Model API ID Best for Notes Gemini 3.5 Flash gemini-3.5-flash Frontier a...
4. What is Gemini 3.5 Flash and what makes it the current flagship model?
Gemini 3.5 Flash (model ID: gemini-3.5-flash ) launched as generally available in June 2026 and is described by Google as delivering sustained frontier performance on agentic and coding tasks. It is the model behind the gemini-flash-latest alias. Key characteristics: Pro-level intelligence at Fla...
5. What are model version types in the Gemini API and which should you use in production?
Gemini model IDs follow a structured naming convention that indicates the model's version type. Choosing the right version type affects stability, rate limits, and billing behaviour. Gemini model version types Type Example Behaviour Production use? Stable gemini-2.5-flash-001 Pinned; does not cha...
6. What is the Antigravity agent and what can it do?
Antigravity ( antigravity-preview-05-2026 ) is Google's general-purpose managed agent released in public preview in May 2026 as part of the Gemini API's Managed Agents feature. It is a fully autonomous agent that runs inside a secure, isolated Google-hosted Linux sandbox container. What Antigravi...
7. What are Managed Agents in the Gemini API and how do they differ from building agents yourself?
Managed Agents is a Gemini API feature (in public preview as of mid-2026) that lets developers build and deploy autonomous, stateful agents that run in secure, isolated Google-hosted Linux sandbox environments. Unlike building your own agent loop, managed agents handle the infrastructure of multi...
8. What is Search Grounding and why is it important for Gemini applications?
Search Grounding connects Gemini models to Google Search, enabling responses based on real-time, up-to-date web information rather than the model's training data alone. This is especially important because all Gemini 3 models have a knowledge cutoff of January 2025. from google import genai from ...
9. What are the Gemini API's multimodal input capabilities?
Gemini models are natively multimodal - they can accept and reason across text, images, video, audio, and documents (PDFs) in a single request. This is a fundamental design characteristic, not a bolt-on feature. Gemini multimodal input types Modality Supported formats Notes Text Plain text, markd...
10. What is the Gemini context window and how do you manage long contexts?
The context window is the maximum number of tokens a model can process in a single request (input + output combined). Gemini models have among the largest context windows of any commercially available AI, with 1 million tokens for most current models. Context window limits by model Model Input co...
11. How does thinking/reasoning work in Gemini models and what is thinking_level?
Gemini 3 series models use dynamic thinking by default - they automatically decide how much internal reasoning to apply before responding, calibrated to the task complexity. Developers can influence this via the thinking_level parameter. from google import genai from google.genai import types cli...
12. What is structured output (JSON mode) in the Gemini API and how do you implement it?
Structured output constrains Gemini to respond in valid JSON matching a developer-defined schema. This makes AI output reliably machine-readable without string parsing heuristics. from google import genai from google.genai import types from pydantic import BaseModel from typing import Literal cli...
13. How does function calling (tool use) work in the Gemini API?
Function calling allows you to declare external functions to the Gemini model. The model then decides when to call them, returning structured arguments you execute in your code. The result is sent back, and the model continues its response using the function output. from google import genai from ...
14. What is the Gemini Files API and when do you need it?
The Files API ( /v1beta/files ) allows you to upload files to Google's servers for reuse across multiple Gemini API requests. It is essential for large files that would be impractical to pass as base64 inline, and for files you want to reference multiple times without re-uploading. from google im...
15. What is the Gemini Live API and what real-time capabilities does it enable?
The Gemini Live API enables low-latency, bidirectional real-time interactions with Gemini models over a WebSocket connection. It supports simultaneous audio streaming in both directions, enabling voice conversations, real-time transcription, and audio-to-audio (A2A) applications. Live API capabil...
16. What are the Nano Banana image generation models and how do they replace Imagen?
Nano Banana is Google's branding for the Gemini native image generation models. They replace the older Imagen models (deprecated, shut down by June 30, 2026) and are natively integrated into the Gemini model family rather than being a separate product. Nano Banana model lineup Model Also known as...
17. What is Veo and what video generation capabilities does the Gemini API offer?
Veo is Google's video generation model family, accessible via the Gemini API. The current production line is Veo 3.1 , with models ranging from high-quality cinematic generation to fast, cost-efficient production of short clips. Veo 3.1 model lineup Model Capability Use case Veo 3.1 (GA) State-of...
18. What are the Gemini API rate limits and how do they differ between free and paid tiers?
The Gemini API enforces rate limits on three dimensions: requests per minute (RPM), requests per day (RPD), and tokens per minute (TPM). Limits vary significantly by model and whether you are on the free tier or a paid billing tier. Key rate limits (mid-2026) Model Free tier RPM Free tier RPD Pai...
19. What is the Gemini Batch API and when should you use it?
The Batch API allows submitting multiple Gemini API requests as an asynchronous batch job, receiving results up to 24 hours later at approximately 50% reduced cost compared to synchronous API calls. It is designed for large-scale, non-time-sensitive workloads. from google import genai from google...
20. How do you use Python and JavaScript SDKs with the Gemini API?
Google provides official SDKs for Python ( google-genai ) and JavaScript/TypeScript ( @google/genai ). Both are available from version 2.3.0 onwards for Interactions API support. # Python SDK setup: pip install google - genai import os from google import genai # Authentication - reads GEMINI_API_...
21. What is context caching in the Gemini API and how does it reduce costs?
Context caching stores a large stable prompt prefix (such as a system instruction, large document, or tool definitions) on Google's servers. Subsequent requests referencing that cache pay a lower price for the cached tokens rather than full input token pricing. from google import genai from googl...
22. What is the Gemini API pricing model and how do you estimate costs?
The Gemini API uses a pay-per-token model. All Gemini 3 models currently in preview have billing enabled; free tiers exist for Gemini 2.5 models. Pricing differs between input tokens, output tokens, and cached tokens. Pricing overview (mid-2026 illustrative) Model Input price Output price Best fo...
23. What is Google AI Studio and how does it differ from Vertex AI for Gemini access?
Google AI Studio (aistudio.google.com) and Google Cloud Vertex AI are two distinct pathways to access Gemini models. Choosing the right one depends on your scale, compliance requirements, and cloud strategy. Google AI Studio vs Vertex AI Aspect Google AI Studio Vertex AI Target user Individual de...
24. What are Gemma models and how do they relate to the Gemini API?
Gemma is Google's family of open-source, lightweight language models built from the same research and technology as Gemini. They are available for free download and can be run locally or deployed on your own infrastructure without the Gemini API. Gemma vs Gemini Aspect Gemma Gemini Open source Ye...
25. How do you implement streaming responses in the Gemini API?
Streaming delivers model output token by token as it generates, rather than waiting for the complete response. This dramatically improves perceived performance in interactive applications and is essential for long responses. from google import genai client = genai . Client() # Method 1: generateC...
26. What is the Python `import antigravity` Easter egg and what does it demonstrate?
The import antigravity Easter egg is a hidden joke built into Python's standard library since Python 3. When executed, it opens a web browser pointing to the XKCD webcomic strip #353, which humorously depicts Python enabling flight by simply installing a library. # Python Easter egg import antigr...
27. What is the Gemini API's approach to safety and content moderation?
Gemini models include built-in safety filters that evaluate both input prompts and output content across four harm categories. Developers can configure the threshold at which content is blocked, balancing safety with utility for their specific application context. Safety harm categories Category ...
28. What is Gemini Deep Research and how does it work as a managed agent?
Gemini Deep Research is a managed agent in the Gemini API (available in preview via Google AI Studio) that performs comprehensive, multi-step research tasks. Unlike single-turn search grounding, Deep Research plans a research strategy, executes many web searches over an extended period, and synth...
29. What is Firebase AI Logic and how does it simplify Gemini integration in mobile and web apps?
Firebase AI Logic (formerly Firebase AI Extensions) is Google's managed backend service that provides a secure, scalable Gemini API integration specifically designed for mobile and web applications. It abstracts away server-side API key management and provides Firebase-native patterns for AI feat...
30. How do you implement multi-turn conversations in the Gemini API?
Multi-turn conversations maintain context across several exchanges. The Gemini API supports two approaches: manually managing conversation history with generateContent , or using the Interactions API's previous_interaction_id for server-side state management. from google import genai from google....
31. What is the Gemini API's text-to-speech capability and how do you use it?
The Gemini API provides text-to-speech (TTS) generation through the Speech API and the Live API . The Speech API generates audio from text using Gemini's native speech synthesis with dozens of available voices. from google import genai from google.genai import types import wave client = genai . C...
32. What are Gemini API embeddings and what models support them?
Embeddings convert text into dense numerical vectors capturing semantic meaning. The Gemini API provides text embedding models for building semantic search, RAG pipelines, clustering, and classification systems. from google import genai from google.genai import types import numpy as np client = g...
33. What is the Gemini API's system instruction and how does it differ from a user prompt?
The system instruction (equivalent to a system prompt) sets the model's persona, behaviour, constraints, and context before any user interaction begins. It differs from a user prompt in that it is processed separately and takes precedence in shaping the model's overall behaviour. from google impo...
34. What is the observable execution steps feature in the Gemini Interactions API?
When using the Interactions API, each interaction object exposes execution steps - a detailed log of everything the model did to arrive at its answer, including thinking steps, tool calls, code execution, and web searches. This provides transparency into model reasoning for debugging and building...
35. How do you handle Gemini API errors and implement robust error handling?
The Gemini API returns standard HTTP error codes. The Python SDK wraps these as typed exceptions from google.api_core.exceptions . Robust applications should handle these systematically with appropriate retry and fallback strategies. import time from google import genai from google.api_core impor...
36. What is Gemini's native audio understanding capability and how do you use it?
Gemini models can directly process audio files - transcribing speech, answering questions about audio content, identifying speakers, and analysing audio characteristics. This is different from the Live API's real-time audio; it processes pre-recorded audio files. from google import genai from goo...
37. What is the Gemini API's grounded generation with Google Search and how does attribution work?
When Search Grounding is enabled, Gemini not only uses real-time web data to improve its response but also returns grounding metadata - source attributions showing which web pages contributed to the answer. This enables you to display proper citations in your application. from google import genai...
38. What are the key deprecated and shut-down Gemini models developers should know about?
Staying current with model deprecations is critical for production applications. The Gemini API has seen significant deprecations throughout 2025-2026, and applications referencing deprecated model IDs will receive errors after shut-down dates. Major deprecations and shut-downs (2025-2026) Model(...
39. How do you build a Retrieval-Augmented Generation (RAG) pipeline with the Gemini API?
RAG enhances Gemini responses with content from your own documents, databases, or knowledge bases. The Gemini API supports two approaches: using the built-in file_search tool (managed by Google) or building your own pipeline with Gemini embeddings and an external vector database. from google impo...
40. What are best practices for building production-grade Gemini API applications?
Moving a Gemini prototype to production requires addressing reliability, cost efficiency, safety, and maintainability. These practices apply across model versions and application types. Production readiness checklist Area Best practice Model selection Use pinned stable model IDs (e.g. gemini-2.5-...
