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

A developer shares another Gemini 4 Argon output that took 13 minutes and about 2% of the weekly quota, calling the model fast and putting the prompt in the comments.

A contributor's celebratory post calls Argon a milestone for the Gemini family, with breakthroughs in reasoning, coding, multimodal understanding and physical intelligence.

A user says Argon is not a benchmark-tuned model, claiming real new SOTA in writing and 3D generation, only slightly behind Opus 5.5 in coding, and shares a video made with it.

A user says Logan reached out by direct message after seeing their posts about Gemini 4 Argon, suggesting the team is watching community discussion.

A user testing Gemini 4 Argon for productivity work says about 10 rounds of conversation exhausted the Pro plan's 5-hour quota, leaving nothing tested.

A user says they subscribed to Google AI Ultra only to find they cannot use Argon 4 for black-box testing.

Blocked by Antigravity's region restriction while trying Gemini 4 Argon, the author shares a tested process for changing a Google account's country to the US and passing review.

A demo shows Gemini 4 Argon teaching a robot to play cornhole just by watching a video.

A user generated a fully articulated 3D clockwork bee with flight mode and exploded view in pure Three.js with no external assets.

A post with a screenshot shows Gemini 4 Argon in the web interface, with the author responding "hmm...." to the rollout.

A developer says Gemini 4 Argon performs well on 3D work and notes that DevinAI is 75% off and worth trying.

A user tests Gemini 4 Argon with a time riddle and says its answer carries a clear big-model signature, sharing a screenshot.