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

Google's official comparison chart shows Gemini 4 Argon topping 14 of 20 rows, with GPT-6 Astra winning 3 and Opus 5.5 winning 2, though the model isn't available for testing yet.

Business Insider reports that Google is internally testing a new Gemini 4 version codenamed Carbon that matches Claude Opus 5.5, possibly an iteration checkpoint of Argon or a new Gemini 4 branch.

LuminaBench says Argon is fully set up in Antigravity and old models are leaving soon, urging Google to release it already.

A joke post likens the still-unreleased Gemini 4 Argon to "the friends we made along the way," poking fun at the release pace.

Business Insider says Google DeepMind is testing three Gemini 4 variants internally, with Carbon showing a big jump in coding and one employee saying it "feels like Opus 5.5"; the public Argon build was internally called Barium-B.

Employees are reportedly testing a newer Gemini 4 version codenamed Carbon that is comparable to Opus 5.5, while earlier Argon versions were compared to Opus 5.

A post jokes that Google's new model codename Argon, implied to be the rumored Gemini 4, is named after a fart.

Employees on the Jetski coding platform say Gemini 4 Carbon's coding performance feels like Opus 5.5, while Argon is still not public.

A user notes that Gemini 3.7 Flash has been removed from Google AI Studio while 3.6 Flash is still available, and asks what is going on.

A Chinese user complains that the releases of Gemini 4, Gemini 4 Flash and Gemini 4 Argon have slipped to next week again, and says English-language X hypes them harder.

A user notices gemini-3.7-flash is no longer available in AI Studio, possibly to make room for Gemini 4 or due to a routing change.

A leak says Google is testing an internal model called Carbon on its coding platform, with one employee comparing it to Opus 5.5 and testers praising its coding, though no public release is confirmed.