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

The VoxelBench Live leaderboard shows Kimi K3.1 Pro at 6th place, currently tied with Gemini 4 Argon.

Post jokes that Gemini 4 Argon, Muse Wonder and Kimi K3.1 Pro all aim to dethrone Claude Opus 5.5, yet each launch ends up as free advertising for Anthropic.

An engagement post asks developers which model they would buy with $100 among Sol 6.1, Kimi K3.1, Opus 5.5 and Gemini 4 Argon.

NovaXCode pushes back on Google's Argon 4, Meta's Muse Wonder and Kimi K3.1 Pro claiming to beat Opus 5.5, calling it benchmark gaming or paid hype and saying Opus 5.5 remains the final boss.

A leak claims the Deep Think mode in the Gemini app is going away soon and will be replaced by Argon, reportedly Gemini 4 Pro.

A user reports Google raised Gemini 4 Argon usage limits about twofold, with Pro members getting roughly $30 worth of Argon access.

A user notes Gemini 4 Argon is out and asks whether anyone has actually tested it on a real codebase yet.

The author measured Gemini 4 Argon's weekly quota rising from 29.46M to about 62.71M tokens, a 2.12x increase worth roughly $10.94 in API value.

A creator runs the same prompt, a 10-second video of a plane landing at sunset, through Kimi K3.1 Pro, Gemini 4 Argon and Fable 5.1 and compares the outputs.

A joke post shows Gemini 4 Argon answering a 'get past the wall' prompt with an absurd chain of rubbing hands sore, splitting the wall and jumping through.

A short take argues Opus 5.5 was the last model not overly benchmark-maxxed, and Gemini 4 Argon may be similar.

The author explains potato mode as one command that fetches only the skill a task needs, like an encyclopedia index; on stream it fanned out to four helpers and included Gemini 4 Argon.