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

TokenGremlin says Gemini 3.5 Pro was promised but never arrived, Gemini 4 Argon was hyped twice without appearing, and Gemini 4 Carbon reportedly is not coming either, doubting any of them will match Opus 5.5 or GPT-6 Astra.

A leaker says Google's latest internal Argon checkpoint is comparable to Claude Opus 5.5 on code tasks, with one employee who tested it saying it "feels like Opus 5.5" but needs more testing.

A leaker says the Gemini 4 Carbon checkpoint matches Opus 5.5, notes Argon still hasn't shipped, and urges Google to release its model.

A leaker says Gemini 4 Argon hasn't launched yet while Google is already testing a new checkpoint called Carbon that matches Opus 5.5.

A user says Google hasn't even released Gemini 4 Argon yet but is already testing something better internally, joking that neither may ever ship.

A demo video shows a reference to Gemini 4 Argon inside the Antigravity CLI, suggesting the codename has reached the toolchain.

The poster notes that Gemini 3.7 Flash has been discontinued only two months after release, which is seen as a sign the Gemini 4 family is taking over.

A user jokes that Google's messaging shifted from "Carbon is our new model" to "Carbon codes like Opus 5.5," and promises a week of promoting Google if Carbon or Argon meets that bar.

A user says Google's internal codenames for the Gemini 4 lineup, Argon, Barium and Carbon, are the best in the industry.

A leak claims Google is about to ship Gemini 4 Argon, that staff are already testing a later checkpoint called Carbon, and that the public Argon is the Barium-B checkpoint rather than the latest one.

Argon has reportedly appeared in Antigravity code with effort sliders and has been registered in Vertex, with a public rollout expected as soon as next week.

A leak says Google is internally testing a new checkpoint called Carbon, said to be approaching the level of Opus 5.5, while Gemini 4 Argon has yet to launch.