As of: 2026-10-14 00:22 CEST.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/v/theoretically-media-testing-gemini-4-argon-800-million-tokens/

# Google Cooked! Testing Gemini 4 Argon & Blasting 800 Million Tokens!

Theoretically Media, 13 October 2026, YouTube. 10,219 views as of 13 October 2026.
Kind: Review.
Watch: https://www.youtube.com/watch?v=DD4oHSH5SEI

## Why it is here

Launch-day hands-on (15:15) by Theoretically Media, a creative-tools channel; the title says the test used 800 million tokens.

## Description (written by Gemini from the video)

**Summary**  
Tim from *Theoretically Media* reviews Google’s newly released flagship frontier AI model, Gemini 4 Argon, after testing it in early access. He examines its performance across real-world projects—including game creation, coding a polyphonic synthesizer, DaVinci Resolve video editing, Skyrim modding, and story architecture—while evaluating its benchmarks, pricing, and 1-million-token context capabilities against competing models.

**What is shown**  
* [00:10] Announcement graphic and branding for Google DeepMind’s Gemini 4 Argon.
* [01:05] Timeline breakdown showing Google’s rapid rollout of four Flash iterations (3.5 through 3.8 Flash) across 106 days prior to Argon.
* [01:36] Google AI subscription tiers interface displaying pricing for Google AI Pro ($19.99/month) and Google AI Ultra ($119.99/month).
* [01:51] Comparative benchmark table evaluating Gemini 4 Argon against GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across coding, reasoning, and cyber benchmarks.
* [02:12] Robotics demonstration video of an Apptronik Apollo humanoid robot playing cornhole powered by Gemini 4 Argon.
* [02:43] Dashboard telemetry of the presenter's personal usage showing 800.45 million total tokens consumed and 6,133 invocations across 11 days.
* [03:05] Gameplay and mechanics of *AI vs Aliens*, an isometric tower defense/civic game developed in Godot with Blender 3D assets generated via Argon.
* [05:00] UI and musical demo of *NEON-84*, a browser-based polyphonic synthesizer, 16-step cyberpunk drum sequencer, and 5-track looper built with Argon.
* [06:18] Gameplay footage of a custom mod importing Doom Guy and weapons into *The Elder Scrolls V: Skyrim*, coded via Argon.
* [07:00] Automated graphics modding in *Skyrim* executed unattended using Argon's `/goal` mode to download and configure ENBs, shaders, and texture packs.
* [07:48] DaVinci Resolve project timeline automated via an MCP connection to Argon, editing a 90-minute interview with Javi López down to a 10-minute cut.
* [09:21] Comparison between an unwieldy Obsidian graph produced by Claude Opus 4.8 and Argon's structured story bible architecture for the sci-fi project *Planet Hell*.
* [10:11] Cinematic preview trailer for the film project *Planet Hell*.
* [12:37] Community showcases comparing Argon with Opus, Astra, and Muse across UI redesigns (Peter Yang), brutalist 3D architectural modeling (Ethan Mollick), and a *Paperboy* game clone (Mike Taylor / Jerrod Lew).
* [14:41] Code leak screenshot referencing upcoming imaging model `gemPix3Pro` (Nano Banana Pro 2).

**Claims & numbers**  
* The presenter states Gemini 4 Argon is Google's flagship frontier model, skipping a mainline 3.5 Pro release after four Flash updates in 106 days.
* The presenter notes Argon is available to Google AI Pro ($19.99/month) and Ultra ($119.99/month) subscribers, without requiring a $500/month plan.
* Gemini 4 Argon launched with an introductory API price of $2 per million input tokens and $10 per million output tokens (with cached inputs discounted by 95%), adjusting on November 30, 2026, to $4/M input and $20/M output (presenter cites announcement text).
* The presenter claims Argon features a 1-million-token output capacity (up from 64k) alongside an agentic continuation protocol for multi-step tasks.
* The presenter notes his personal testing racked up 800.45 million total tokens, 730.49 million cached tokens read, 8.09 million output tokens, and 6,133 invocations in about a week across desktop and Mac setups.
* Citing a discussion with Google executive Josh Woodward, the presenter claims Google fine-tuned Argon specifically to eradicate common AI writing clichés such as "it's not X, it's Y," "delve into," and "rich tapestry."

**Notable quotes**  
* [00:29] "Basically answer the question: did Google cook?"
* [01:21] "Skipping out on 3.5 and going directly to 4, surnamed Argon."
* [12:06] "One thing I will definitely say about that million-token context is that it does feel very roomy."

**Assessment**  
This is a hands-on review and workflow analysis of Gemini 4 Argon by a creator who tested early-access builds across software development, music production, game modding, and creative writing. All showcased tools and projects represent live, functioning demonstrations in Godot, Blender, DaVinci Resolve, and game engines rather than simulated mockups.

_Described by gemini-3.8-flash on 2026-10-13 from the video's audio and frames._

## Related

- 2026-10-13: [Google releases Gemini 4 Argon broadly](https://postcutoff.com/e/2026-10-13-gemini-4-argon-general-availability/)

## People in it

- [Ethan Mollick](https://postcutoff.com/person/ethan-mollick/), Associate professor, Wharton School