Google's AI Stack: Gemini, Antigravity and Project Astra

Google has more AI names than anyone, and they live on every layer of the map: a model family (Gemini), products (the Gemini app, Gemini Live), agents (Gemini CLI, Antigravity) and a research prototype (Project Astra). Let us take them in order.

Gemini: the model family

Gemini is Google's family of models. The word is Latin for "twins", and it is commonly linked to the 2023 combination of Google's two AI labs, Google Brain and DeepMind, into one team. The family is organised by tier, in the same way as Claude and GPT:

Versions count upward: at the time of writing Google's pricing page lists Gemini 3.8 Flash as the latest Flash, and Gemini 3.1 Pro as a preview of the Pro tier. Gemini models are natively multimodal (text, images, audio, video) and have very long context windows, which is one of their signature strengths.

Using Gemini: four doors

1. The Gemini app and Gemini in Google products (Gmail, Docs, Sheets), on the free plan or a paid Google AI plan.
2. Gemini Live, a voice and camera mode that descends from Project Astra (more below).
3. The Gemini API for developers, through Google AI Studio or Vertex AI.
4. Agents: Gemini CLI and Antigravity.

Subscription versus API

Subscription: Google AI plans. The consumer ladder, from the Google AI plans page, is AI Plus, AI Pro and AI Ultra. Each raises your Gemini usage (roughly 2 times, 4 times and up to 20 times the free access), adds storage, and Ultra adds Deep Think reasoning and higher limits in the agents Antigravity and Jules. The plans bundle other Google perks too, so the price is not only for AI. Prices depend on your country, so read the page rather than a screenshot.

API: per token, with a free tier. On the Gemini API pricing page the Flash-class models have a free tier and a paid tier, and Pro models are paid. As a snapshot: Gemini 3.8 Flash is listed at $0.75 per million input tokens and $3.75 per million output tokens through the end of 2026 (higher after), Gemini 3.5 Flash-Lite at $0.30 and $2.50, and Gemini 3.1 Pro Preview between $2 and $4 for input and $12 to $18 for output, depending on prompt length. The difference between free and paid is not only money: the free tier has lower rate limits, and Google may use free-tier data to improve its products, while the paid tier does not.

The free API tier is one of the most generous ways for a student to experiment with a frontier-style model, with the trade-off that you should not send private data through it.

Gemini CLI: Gemini in your terminal

Gemini CLI is an open-source (Apache 2.0) terminal agent.

npx @google/gemini-cli

Three ways to sign in:

• Google account login, which needs no API key and, per the project's README, gives a free allowance (60 requests a minute and 1,000 a day) with a one-million-token context window.
• A Gemini API key, for control over the model and paid upgrades.
• Vertex AI, for enterprise teams with billing and higher limits.

It supports the Model Context Protocol (MCP), search grounding, file operations and shell commands, and a GEMINI.md file that tells it about your project, the same idea as CLAUDE.md.

Antigravity: Google's agent-first development platform

Antigravity is Google's agentic development platform, free for individual developers with an enterprise version for organisations. Its main pieces, from the Antigravity site:

• A command center (Antigravity 2.0) for running several local agents in parallel, grouping them by project and automating tasks. This "manager" view is the defining idea: you supervise agents instead of typing every edit.
• An IDE, a full editor with agent management and codebase understanding.
• A terminal-first CLI.
• An SDK in Python, for prototyping your own agents.
• Extensions.

Which models it offers, and what the free allowance is, has shifted since its launch in late 2025, so read the current page instead of relying on a review. The name is a wink: code that feels like it is floating, with agents doing the heavy lifting.

Project Astra: the prototype behind Gemini Live

Project Astra is a Google DeepMind research prototype exploring "a universal AI assistant". It is not something you can buy. Its goals, per the DeepMind page:

• Natural interaction: fast, low-latency voice across languages.
• Proactive help: it can start a conversation instead of only replying.
• Action: using tools such as Search, Gmail and Calendar on your behalf.
• Memory: remembering your preferences across sessions.
• Multimodal understanding: seeing through a camera, listening, reading.

Its capabilities are being folded into Gemini Live and tested in Search and in prototype glasses, and a version built with the blind and low-vision community is available to trusted testers. The word astra is Latin for "stars", a fitting name for a reach-for-the-horizon project. The practical takeaway: Astra is where you look to see what Google assistants will do next, and Gemini Live is where some of it already works today.

What you can do with Google's stack

• Ask questions about a long PDF, video or codebase thanks to the long context.
• Talk to your phone camera with Gemini Live.
• Write, research and summarise inside Gmail and Docs.
• Build apps for free with AI Studio and the API.
• Delegate coding to Gemini CLI or several Antigravity agents at once.

Links

• Gemini API pricing: ai.google.dev/gemini-api/docs/pricing
• Google AI plans: one.google.com/about/google-ai-plans
• Gemini CLI: github.com/google-gemini/gemini-cli
• Antigravity: antigravity.google
• Project Astra: deepmind.google/project-astra