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

Developer Ammaar used Gemini 4 Argon to build a runner game where you fly a Tesla Roadster over traffic, barrel roll through the sky, and collect air power-ups for a high score, with a playable demo and code.

A read of the Gemini 4 Argon Minecraft experiment finds four agents with no assigned roles finished the task after 27 hours, 2,783 programs, and 605 deaths, which the author sees as persistence becoming a shared core capability of Opus 5.5, GPT 6 Astra, and Gemini 4 Argon.

A Portuguese-speaking user says Google has released Gemini 4 Argon on Antigravity's Pro plan and asks whether it is worth trying.

A user says four prompts on a 20-page PDF used 91% of the 5-hour limit on Gemini 4 Argon, and questions whether upgrading from Pro to Ultra is worth it.

adamdotnew says Gemini 4 Argon performed best in its CAE simulation tests and is now Adam's recommended model for anyone running simulations.

A user asks whether discussion around Gemini 4 Argon has cooled off, prompting debate about interest in the rumored Gemini 4 Pro model.

A screenshot post shows the suggestions Gemini 4 Argon gave on how to turn your life around in six months.

Arena's official weekly report says Gemini 4 Argon (High) ranked #5 in Agent Arena with a 13.2% net improvement, alongside Xiaomi's MiMo and Meta's Muse models.

A user reports that Gemini 4 Argon is available in the Antigravity CLI for the sub-$20 Google AI Pro plan, including discounted subscribers but not free users.

Gemini 4 Argon has been added to ChessBench, which will switch to a new 1-100% scoring system and later let users play the models directly to measure real Elo.

A Google DeepMind paper says four Gemini 4 Argon agents with no assigned roles wrote 2,783 programs, died 605 times and killed the Ender Dragon in 27 hours, the first fully autonomous clear of its kind.

The author gave Gemini 4 Argon and Codex Pro the same task of building a playable Angry Birds; Codex finished 16 seconds faster but needed another 2m49s to fix its broken slingshot, and Gemini 4 ran through MixRoute.