How to Use Gemini 4 Argon: Every Way to Access Google's New Frontier Model

Before anything else: match your account to an entry point

1. 1

   Match your plan to an Argon entry point

   Google's launch post lists five doors: the Gemini API paid tier for developers, Gemini Enterprise for businesses, the Gemini app and AI Mode in Search for Ultra and Pro subscribers, plus application programs for scientists and cybersecurity defenders. Find the row that matches your account before you touch anything else.

   

2. 2

   Note the introductory price before you build

   The same post states the launch price: $2 per million input tokens and $10 per million output tokens, with cached input tokens billed at 5%. Bijan Bowen's walkthrough shows the pricing docs putting the offer at $2 and $10 through November 30, 2026, doubling to $4 and $20 after that.

   

Entry point 1: the Gemini API for developers

1. 3

   Read the gemini-4-argon model card

   Developers start at ai.google.dev. The model page lists what matters before your first call: a 1,048,576-token context window, 65,536 default output tokens (up to 1,048,576 with long decode), and text, image, video, audio, and PDF inputs. Bijan Bowen walks this exact page in his launch video.

   

2. 4

   Or call the model through OpenRouter

   BridgeMind's stream shows the google/gemini-4-argon listing: $2 and $10 per million tokens at 50% off, a 1.0M context window, and Google Vertex plus Google AI Studio as providers. If you already route requests through OpenRouter, adding Argon is a one-line change.

   

Entry point 2: Antigravity for agentic coding

1. 5

   Switch the model and effort in the Antigravity CLI

   BridgeMind runs Argon inside Antigravity on an Ultra account. The CLI header reads Gemini 4 Argon (Medium) by default, and the /effort command opens the reasoning-effort picker - medium is the default, high costs more tokens. The status bar confirms the active pair, like Gemini 4 Argon · medium.

   

2. 6

   Start a desktop conversation on Argon High

   In the Antigravity desktop app, Paul J Lipsky opens a new conversation and the model chip already reads Gemini 4 Argon High. Pick a project - his sidebar holds Capture the Flag, Space Fighter, and Rocket Launch - then describe what you want built; the app writes and manages the code for you.

   

3. 7

   Invoke /boost for multi-agent tasks

   Typing a slash in the Antigravity prompt box opens the command menu, where boost is described as the multi-agent orchestrator for complex tasks - Bijan Bowen selects it right before launching his 3D skate-game build. The docs note /boost works on paid plans across Antigravity 2.0 and the CLI, while teamwork-preview targets longer-horizon projects.

   

4. 8

   Watch the weekly and five-hour quota

   BridgeMind checks Models and Quota in the CLI: the Gemini group shares a weekly limit and a rolling five-hour limit, consumed in proportion to token cost. Ten minutes of launch-day testing left 97.32% of the week and 92.83% of the five-hour window on his Ultra plan - large builds drain it much faster.

   

Entry point 3: the Gemini app for everyday use

1. 9

   Confirm Argon High in the Gemini app

   In the Gemini app the composer's model dropdown reads Argon High - Paul J Lipsky uses it for a horror story sized at 20 to 30 minutes of reading, delivered as a Canvas document you can export. Google is rolling Argon out to AI Pro and Ultra subscribers gradually, so the chip may take a few days to reach you.

   

2. 10

   Ask for research and a deck in one prompt

   Lipsky's own launch video was illustrated by Argon: he asked for a slide deck about the model and received an 8-slide, 16:9 PPTX with a slide-by-slide breakdown and an Open button. Requests like this also run inside Gemini Spark, where Lipsky says the Argon upgrade is most visible.

   

What Argon actually did on launch day

1. 11

   See where Google aims the model

   Google's own launch graphic puts Gemini 4 Argon at the center of coding, finance, creative writing, and legal. The three launch-day videos then test code, games, and security review up close, plus long-form writing in the Gemini app.

   

2. 12

   Read the whole benchmark table, not just the wins

   The official table shows Argon first in 13 of 19 benchmarks with one tie, leading knowledge-work rows like Vals Index and Harvey's Legal Agent Benchmark against GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5. Note the rows it does not lead - Terminal-bench 4.0 and PostTrainBench still favor Opus 5.5.

   

3. 13

   Build a playable game in two prompts

   Lipsky's first Argon build in Antigravity was Capture the Flag: Chroma Arena, a top-down paintball game - one prompt created it, and a second added the AI difficulty slider in the top-left corner. He says the first pass looked plain until he pushed for more visual direction in the prompt.

   

4. 14

   Audit code you didn't write

   Lipsky pointed Argon at an Apple Watch app he had built with a different model and asked it to find bugs. About six minutes later it returned a prioritized list inside the Agent Talkie project - Priority 1 covered a hidden fallback-bot name that could block Send to Watch.

   

5. 15

   Let it grade the risk it finds

   The scariest finding got its own visual: a webhook URL marked exposed above a gauge reading low risk. Lipsky explains that a leaked webhook would have let a stranger command his Grok Bot, and says Argon found the bug and helped him patch it - a security-review workflow worth copying.

   

Frequently asked questions