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

A report says Google is internally testing a model called Carbon before Gemini 4 Argon is even public, with one employee describing its coding feel as like Opus 5.5.

A user noticed Gemini 3.1 Pro, 3.6 Flash and 3.7 Flash are all marked Leaving soon, and guesses Google may be preparing a public rollout of Gemini 4 Argon.

A leak says Argon has been added to Antigravity with only low, med and high reasoning levels, plus 256k, 512k (1.3x) and 900k (1.8x) context tiers priced separately.

A report says the newly released Gemini 4 Argon is seen by internal staff as underperforming, and DeepMind's new lead hinted the next generation may not arrive until October.

A user noticed the Gemini web interface has been redesigned and wonders whether it signals Gemini 4 launching the same day.

A leak claims Google is internally testing a Gemini 4 checkpoint called Carbon whose coding ability rivals Opus 5.5, and speculates it is Gemini 4 Pro while Argon is 4 Flash.

A poster says more people suspect Gemini 4 Argon will never launch, noting Carbon was tested first and that no trusted benchmarks show either model beating Opus 5.5 or GPT 6 Astra.

A user shares a screenshot saying Gemini 4 Argon is unbeatable in that scenario, with an image showing the result.

An image claims Argon is already built into Antigravity, which the poster reads as a sign it could launch any day.

An AI stock daily says Google will release Gemini 4 Argon with a 1M-token output limit and is already testing a stronger Carbon version aimed at Opus 5.5.

A Chinese user complains that Gemini 4 Argon still hasn't launched while Google is already pushing an internal Gemini Carbon said to match Opus 5.5.

The author ranks Opus 5.5 and GPT-6 Astra at the top and Gemini 3.8 Flash in the middle, says Claude now beats GPT across the board, and looks to Gemini 4 Argon to change the game.