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

A leaker says the latest benchmark results show Gemini 4 Argon clearing Claude Opus 5.5 and GPT Astra 6 effortlessly, with a leaderboard image attached.

The poster lays out a timeline from Gemini 3.1 Pro on Feb 19 to Gemini 4 Argon on Sept 30 and argues competition is keeping Google from shipping Pro-tier models.

A rumor says Google employees find Gemini 4 Carbon comparable to Opus 5.5 for coding, while Argon is not yet widely available.

The post says Gemini 4 Argon scored 10 points on a new benchmark while OpenAI, Anthropic, Meta and xAI all scored zero.

Reddit and X users claiming Ultra and Pro access have posted screenshots of Argon in the model picker, but some suspect the images are AI-generated and no real chat demo has appeared.

Google employees are reportedly testing a new Gemini 4 checkpoint called Carbon on the internal coding platform Jetski, with coding ability said to rival Opus 5.5; the public Argon release is internally codenamed Barium-B.

A new benchmark screenshot claims Gemini 4 Argon leads the rankings, but the post gives no specific tests or scores.

A screenshot claims the latest Gemini 4 Argon benchmark results lead Claude Opus 5.5 and GPT-6 Astra by a wide margin, described as a one-sided win.

A rumor says Google is clearing space for the full Gemini 4 lineup: Argon is reportedly configured in Antigravity, older models will leave in days, and Argon is said to be the Pro model with Gemini 4 and Gemini 4 Flash possibly launching alongside it.

A user found low, medium and high configurations for Gemini 4 Argon in Antigravity, suggesting a public release is close.

A post with an image in which the author says "thank you gemini 4 flash" to express satisfaction with the output; the details are in the image.

Google employees are testing a new Gemini 4 version called Carbon internally and say it codes like Opus 5.5, while Gemini 4 Argon, released October 1, is still limited to security whitelist customers.