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

A user says Gemini 4 Argon has strong vision and spatial understanding, iterating in Blender until it produces a looping animation.

The author found that Claude, without being asked, automatically used Gemini 4 Argon as an audio-understanding proxy, and wonders whether it can have auditory taste.

Google Research released Deep Review to the public, a multi-agent system that reviews papers, reports, and legal and financial documents, with two free reviews per week and support for Gemini 4 Argon.

Blazerail is giving away 2 million free tokens so developers can start testing Gemini 4 Argon, claimable on its website.

Given the same prompt to animate a Pixar-style seaside town being flattened by a tsunami in 15 seconds, Opus 5.5 delivered a complete scene while Gemini 4 Argon mostly washed the screen out to white.

The author used Gemini 4 Argon's coating and material capability to generate a jet, with a video showing the result.

The author criticizes paid hype in AI, says Muse Wonder Full felt strange to try, and argues Gemini 4 Argon was the week's real standout launch, calling for more builders to share honest tests.

Google's Josh Woodward says Argon targets AI slop, listification and stock phrases, using professional writers' markup as golden training data for more concise, precise output.

A user complains that Gemini 4 Argon's quota is used up in under half an hour and calls for the 5-hour limit to be removed, comparing it to a stripped-down GPT.

With the same prompt and /goal, Gemini 4 Argon completed the task in 38 minutes versus 1h 25m for Astra and 2h 10m for Fable 5.1.

A post says 2 million free tokens for trying Gemini 4 Argon can be claimed at blazerail.com, with a video demo attached.

A user jokes that Google, despite having Gemini 4 Argon, did not vibe code its own link-in-bio page and instead used Linktree's free plan.