Google Just Caught Up (Gemini-4 Argon)

Summary
Matt Wolfe reviews Google’s newly released flagship AI model, Gemini 4 Argon, covering its technical specifications, benchmark placements, pricing, and integration into tools like Google Antigravity and the Gemini web interface. He runs hands-on evaluations spanning SVG generation, browser game creation, physics simulation, multi-constraint reasoning, 3D printable STL modeling, and agentic workflows, concluding with a look at its application in robotics.

What is shown

• Benchmark & Architecture Overview [01:47]: Benchmark tables comparing Gemini 4 Argon against GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5, highlighting its 1M output token capacity.
• Artificial Analysis Index [03:23]: Leaderboard standings showing Gemini 4 Argon scoring 53 on the Intelligence Index and its cost-efficiency comparison.
• BuseyBench SVG Test [08:26]: Vector art generation benchmark depicting an SVG illustration of Gary Busey.
• Game Generation in Antigravity [09:34]: A 3D browser clone of Megabonk generated in one shot using Gemini 4 Argon in Google Antigravity, including character selection and playable third-person arena combat.
• Physics Playground Web App [10:51]: An interactive projectile motion simulation with trajectory prediction, adjustable gravity presets (Earth, Mars, Moon), wind speed, and analytical validation.
• Gemini Chat Reasoning & Research [11:58]: Solving a complex multi-task, constraint-based project schedule (Alex and Blair) and producing a comparative table of subscription terms across consumer AI plans.
• 3D Modeling & Printing [14:07]: Model-generated gary_busey_bust.stl inspected and sliced in Bambu Studio, followed by Wolfe displaying the physically 3D-printed purple plastic bust on camera [14:44].
• Gemini Task Mode (Agentic Slides) [15:10]: A multi-step presentation prompt resulting in a completed 6-slide Google Slides deck about prompt injection, despite the agent UI stalling during execution tracking.
• Robotics Cornhole Demonstration [16:37]: Video footage of an Apollo humanoid robot utilizing Gemini 4 Argon to observe human demonstrations and successfully throw cornhole bags into a board target.

Claims & numbers

• Context & Output: Gemini 4 Argon supports an industry-leading output token limit of 1 million tokens (up from 64k tokens), which Wolfe notes equates to roughly 750,000 words [02:42].
• Benchmarks cited: On Vals Index, Gemini 4 Argon scores 68.9%; AutomationBench: 51.3%; DeepSWE v1.1: 77.9%; Vibe Code Bench: 91.9%; Terminal-Bench 4.0: 57.4% [01:47]. On Artificial Analysis Intelligence Index, it scores 53 [03:29].
• Pricing:
  • API introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off [07:33]. After the introductory period ends in late November/December 2026, prices increase to $4/M input and $20/M output [08:03].
  • Consumer access requires Google AI Pro ($19.99/month) or Google AI Ultra (from $99.99/month) for the Gemini app and Search AI Mode, while Antigravity access specifically requires the Google AI Ultra plan [07:09].
• Robotics Performance: The Apollo humanoid robot running Gemini 4 Argon managed to sink cornhole throws 7 times in a row after observation and simulation practice [17:11].

Notable quotes

• [02:43] “No other model is giving you a million tokens of output.”
• [14:44] “Yeah, it actually did print my Busey bust.”
• [19:15] “One reason I never ever count Google out is because Google owns such a huge ecosystem.”

Assessment
This is an independent hands-on product review combining benchmark analysis with real-time and pre-computed demonstrations. All software outputs—including the 3D game, physics simulator, generated Google Slides deck, and a physical 3D print—are verified on screen, though the presenter notes that Gemini’s multi-step “Task” agent interface experienced UI stalls and slow execution during early-access testing.

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