Gemini 4 Pro / Flash: benchmarks, hands-on tests and leaks

A weekly live program discussing Gemini 4 Argon, Muse WonderFull, memory features and the AI stock selloff.

A satirical post in which the author tries Argon at a Google Store, asks whether it is an Opus killer, and concludes it is 40 tps benchmark filler.

The author gathers a thread of posts about using Gemini 4 Argon with Antigravity, rounding up recent hands-on tests and discussion.

A user posts that Gemini 4 Argon performs strongly on cybersecurity and offensive tasks, an unofficial assessment of its capabilities.

A developer rebuilt their portfolio page in the visual style of jobwhisper using Google's new Gemini 4 Argon and shared a screen recording.

A Japanese morning briefing says Gemini 4 Argon is now public, its API is half price until the end of November, AI Studio's free tier can't use it, and it's rated on par with Opus 5.5.

On the same prompt, Argon finished in 3 minutes for $0.5 while GPT-6 Astra ran 25 minutes and cost $11, with hallucination rate and limited-time discount details also cited.

TestingCatalog reports that Google announced Gemini 4 Argon is available through the paid Gemini API and the Gemini platform for developers and enterprises, plus Gemini for Ultra subscribers.

A user says Argon did not work on the Pro tier and only became usable after upgrading to Ultra in Google's Antigravity 2.0 desktop app, and that the early experience feels little different from Claude Code.

A Japanese user compiled Gemini 4 Argon's full 3D model build process into a single illustrated note for anyone who wants to learn the workflow.

The author asked Gemini 4 Argon to create a 3D city with a first-person iPhone walking perspective, with a video demo.

A Japanese post says Argon's highlight is a 1 million token output limit (not input), enough to build high-quality 3D in Blender in one go, and that Google Cloud claims frontier-level performance on complex workflows in software development, knowledge work and cyber defense.