Gemini 4 Argon Tested: Almost Opus 5.5 for a Quarter of the Price

Google just opened Gemini 4 Argon to everyone, so I put it through my VisualForge benchmark: the same interactive 3D web pages I give every model, each one opened, clicked and scored in a real browser. In this video: • what Google released: announced on 30 September, open to everyone on 13 October (Gemini API and AI Studio, the Gemini app, AI Mode in Search, Antigravity) • Google's claims about code, security and science, with clips from Google's own videos • Google DeepMind's benchmark table, checked row by row against GPT-6 Astra, Fable 5.1 and Opus 5.5, including the 1 million output tokens fine print • my four tests: Aero Lab, Jet Odyssey, Blast Island and Launch Day, with Opus 5.5 and Sonnet 5.5 for comparison • my results: 7.22 out of 10, third of 17 models on the same four tests, behind Opus 5.5 (7.45) and GPT-6 Astra (7.36) • cost: about $1.43 a test, against $6.30 for Opus 5.5 and $5.37 for GPT-6 Astra. Argon writes about half as many output tokens as Opus (about 111k a test against 229k) • price and how to get it: $2 in / $10 out per million tokens until 30 November, then $4 / $20, the same as Opus 5.5 Settings: Gemini API (gemini-4-argon) at high effort, the same effort as Opus 5.5, GPT-6 Astra and Sonnet 5.5 on my board. One run, in release week. Argon got six tests and two did not finish: on Camera Shop the stream from Google cut off after 48 minutes, and the wildcard hit the monthly spending cap on my Google API account. So every model here is compared on the same four tests that did finish. Google's numbers are labelled as Google's claims. Costs are estimates from each run's token counts at API prices (checked 13 October 2026). Every page Argon built, with its scores, tokens and cost: https://visualforgebench.com/run/?id=20261013T193341_gemini-4-argon_81c6 All results: https://visualforgebench.com (the site's leaderboard scores every page a run made, so its ranking can differ slightly from the same-four comparison in this video) Chapters 0:00 Intro 0:32 What Google released 1:11 Google's claims 2:26 Google's benchmarks, checked 3:38 The 3D claim 4:34 My benchmark 5:27 Aero Lab 6:12 Jet Odyssey 6:55 Blast Island 7:35 Launch Day 8:58 My results 9:50 Price and how to get it 10:44 Verdict Sources (read 13 October 2026) - Google, Gemini 4 Argon announcement (30 September): https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/ - Google, Gemini 4 Argon for everyone (13 October): https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/ - Google, Gemini app availability: https://blog.google/innovation-and-ai/products/gemini-app/gemini-4-argon-availability/ - Google DeepMind model page and benchmark table: https://deepmind.google/models/gemini/ - Gemini API pricing: https://ai.google.dev/gemini-api/docs/pricing - Gemini API model page: https://ai.google.dev/gemini-api/docs/models/gemini-4-argon - Google AI plans: https://gemini.google/subscriptions/ - OpenRouter: https://openrouter.ai/google/gemini-4-argon - Google DeepMind, Chromium codebase maintenance: https://www.youtube.com/watch?v=LVEfFyijQ2c - Google for Developers, 2D concept to 3D cathedral: https://www.youtube.com/watch?v=OYp5j2RihPE - Google DeepMind, humanoid robot cornhole: https://www.youtube.com/watch?v=9PICEmtfpWU