Gemini 4 Argon - Out and Tested!!

Summary
Sam Witteveen reviews and tests Google DeepMind’s newly released frontier model, Gemini 4 Argon. He demonstrates interactive multimodal coding applications generated by the model, evaluates its benchmark standing against rival frontier models, and explains Google’s broader strategy for positioning the model across its enterprise and consumer platforms.

What is shown

• [00:05, 00:31, 00:57] Interactive 3D Five-Story Pagoda Sanctuary (“Gojo-no-Tō”): A browser-based Three.js/WebGL interactive scene with camera presets and dynamic day/sunset/night lighting controls.
• [00:36, 02:20] “RB-338 Acid Rack Studio”: A browser synthesizer and drum machine inspired by Propellerhead ReBirth, generated from a prompt in the Gemini web app, featuring playable TB-303 synth engines, TR-808/TR-909 sequencers, and real-time audio playback [02:28].
• [00:40, 05:53] “Culinary-Ballistic Hammer-Whisk Schematic”: An interactive technical blueprint UI with exploded part views, CAD callouts, and aerodynamic force diagrams.
• [03:11] Launch showcase reels displaying Argon-scripted Blender 3D animations and audio piped to a vintage CRT monitor, alongside an intricate procedural Japanese temple garden.
• [05:25] Artificial Analysis Intelligence Index chart comparing frontier model scores.
• [05:46] Embodied Reasoning Question Answering (ERQA) benchmark graph comparing leading models.
• [06:27] Presentation footage from the Google for Business keynote highlighting customer token consumption metrics.

Claims & numbers

• The presenter says Gemini 4 Argon was first announced on September 30, 2026, and is now generally available across the paid Gemini API, Google Antigravity, Google AI Pro and Ultra tiers in the Gemini app, and AI Mode in Search [00:48, 01:05].
• The presenter claims Argon is the largest model Google has pre-trained to date and marks the launch of the Gemini 4 family [04:26, 04:42].
• On the Artificial Analysis Intelligence Index, the presenter reports Gemini 4 Argon scores 53, tying GPT-6 Astra (53) and Claude Fable 5.1 (53), behind Claude Opus 5.5 (58), Claude Sonnet 5.5 (56), and Meta’s Muse Wonder (56) [05:25].
• On the Embodied Reasoning Question Answering (ERQA, Tool off) benchmark, the presenter shows Gemini 4 Argon leading at 87.0%, ahead of Claude Opus 5.5 (82.3%), GPT-6 Astra (81.0%), and Claude Fable 5.1 (74.0%) [05:46].
• The presenter notes Google reported that approximately 500 enterprise customers have each processed over 1 trillion tokens on Gemini [06:33].
• The presenter states that leveraging the 1M output token window requires a specific continuation protocol flag in the Gemini API and is not available in the consumer web app [08:58].
• The presenter states promotional pricing is $2 / million input tokens and $10 / million output tokens (discounted 50% from $4 / $20) through December 1 [10:30].

Notable quotes

• “This is a really, really, really good model. But is it the best model out there? I actually don’t think it is.” [00:14]
• “For me the stand out things here is anything that combines sort of visuals, code, and media in one kind of task.” [01:36]
• “It’s that loop of sort of code, render, look, fix something up that continually seems to be doing well here.” [04:14]

Assessment
This is an independent hands-on review and strategic analysis. The presenter shows real, functional web applications generated using Gemini 4 Argon while providing a balanced evaluation that acknowledges its multimodal code-generation strengths without overhyping its standing against dedicated coding agents like Claude Code or Codex.

Described by gemini-3.8-flash on 2026-10-13 from the video’s audio and frames.