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

A user complains that it has been ten days waiting for Gemini 4 Argon, free users lost access to the Flash model the same day, and the Gemini App added thinking_level with an odd Chinese translation.

A Chinese user jokes that the wait for Gemini 4 Argon has been longer than Musk's Grok hype, calling it a "futures model."

Argon isn't widely available yet, and now there are signs of Gemini 4 Flash, leading the poster to guess Google is preparing the whole Gemini 4 lineup at once.

Business Insider reports Google is internally testing Carbon on its Jetski coding tool, with staff saying its coding "feels like Opus 5.5," though there are no public benchmarks or release date.

A user jokes that Google has solved a world-class problem: keeping its products always ahead while never letting users actually try them, as Carbon is tested before Argon launches.

A satirical comparison says OpenAI lets users experience AI, Anthropic makes them follow rules, and Google leaves it to users to imagine how strong Gemini 4 really is.

A poster says Gemini 4 Argon is barely available while Google is already testing Gemini 4 Carbon internally, with employees calling its coding ability close to Claude Opus 5.5.

A Japanese news flash citing Business Insider says Google is internally testing a new Gemini 4 version codenamed Carbon, described as "like Opus 5.5" at coding and newer than the unreleased Argon.

The Gemini product lead and a Google Cloud AI product manager both say Argon is coming, likely available to Pro users, and the poster looks forward to its speaking style more than its coding ability.

TestingCatalog found hidden low, medium and high reasoning tiers for Gemini 4 Argon in Antigravity, while Gemini web now shows the same tiers for all models; the post also cites a Business Insider report on an internal checkpoint called Carbon said to match Opus 5.5 on coding.

Citing Business Insider, the post says Carbon is live on Google's internal coding platform Jetski, where employees compare its coding ability to Opus 5.5, and describes the codename chain as Argon, then Barium, then Carbon.

The user expects Gemini 4 Argon to ship publicly first, followed by Carbon, which is said to match Opus 5.5, and says they are more interested in Carbon.