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

A test claims Gemini 4 Argon spent 33 minutes in Antigravity generating a multi-island voxel pagoda entirely in code, including visuals, scenes and audio.

A short post says the Gemini 4 Argon comeback is personal.

A Thai post says Gemini 4 Argon is now available on Google Cloud, aimed at long-context multi-step tasks, with limited-time pricing of $2/$10 per million tokens.

A tester gave Argon misleading API docs with invented methods; it caught most fakes and suggested alternatives, but still confidently used one of them.

A joke video calls Gemini 4 Argon "sucks" and also mocks Muse Wonder as worse than Opus 5.5.

Google employee Logan Kilpatrick replies that many teams across Google worked on the model and that the Gemini 4 era is only starting.

A user says Gemini 4 Argon is out and works on both of their Google AI Pro accounts, and asks whether GPT can be dropped now that Gemini 4 has arrived.

A user tests the paid mobile Gemini 4 Argon tier, citing gains in visual and spatial reasoning, quantitative modeling and long-document cross-checking, plus Flow 3 video tools and guided learning.

An AI safety account says Gemini 4's safety testing gave attackers a "substantial capability uplift," eliminating months of trial-and-error research on biological weapons.

A user built a multi-island voxel pagoda 3D scene with Gemini 4 Argon in Antigravity in 33 minutes, with audio generated purely in code, and notes Google's demo used an agent loop that repeatedly critiqued its own output.

A user tests Gemini 4 Argon's Chinese writing by asking for an original O. Henry-style short story and posts the full text of "The 49th Bad Review" for feedback.

A short video jokes about how developers' working lives changed after the release of Gemini 4 Argon, part of the launch-day community reaction.