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

The author argues that many Google and DeepMind staff leaving to found AI startups explains Gemini/Argon's weaker results, sharing two ChatGPT-compiled lists.

GitHub Copilot CLI shipped one release with six improvements, headlined by Gemini 4 Argon command hook support plus /ask enhancements.

A timeline post recaps seven days of AI releases, including Google's Gemini 4 Argon launch and OpenAI rolling GPT-6 out to all ChatGPT users.

Claude Opus 5.5 and Gemini 4 Argon each built a rotatable fluffy 3D watermelon for $8.98 vs $0.34 in API cost; the author prefers Opus's fur but calls Gemini impressive for the price.

Open-source whiteboard app Excaliflow gains multi-canvas tabs and auto-save, which the author says was built with Gemini 4 Argon in a single Antigravity session.

A one-line prompt had Gemini 4 Argon build a minimal UI library of realistic physics-animated toggles, and the author praises its micro-animations and design taste.

A Turkish-language joke comparing doctor shortages to Gemini 4 Pro, with no substantive Gemini 4 information.

Coinbase Predict's weekly news list includes Google officially launching its most powerful model, Gemini 4 Argon, alongside Apple stock and Fed data items.

A Japanese user notes Gemini 4 Argon is now usable in Antigravity but finds the IDE clunky, and shares heavy image generation with Dot.

A personal tier list places Gemini 4 Argon in T0 as the go-to for every task with price as the only constraint, tiering it against Opus 5.5 and Gemini 3.7/3.8 Flash.

A user reports Google AI Ultra 20X's Gemini 4 Argon quota rose 75–90% after reset, to roughly 650M tokens per 5 hours and over 2.5B weekly, outlasting GPT 20X.

A video comparison claims Gemini 4 Argon completes an Eiffel Tower task for $0.23 while Claude reportedly spent $33.45, framing it as cheap, fast and good enough.