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

A JIO account with 18 months of Pro was included in a gray test of Gemini 3.8 flash-exp-a, which generated a 3D aircraft carrier from a blueprint line drawing, and the author calls the result fairly advanced.

A user with an 18-month Pro account in the gray test used Gemini 4 Argon in one-shot mode to produce a 3D showcase of a rainy-night convenience store in an anime style.

A Pro account in the gray test generated a miniature railway town with Gemini 4 Argon in one-shot mode, which the author rated as close to Opus 5 level.

A user with an 18-month Pro account in the October 10 evening gray test had Gemini 4 Argon generate the Chinese classical building Taihe Yunque in one-shot mode, with striking results.

Business Insider reports an employee saying Gemini 4 Carbon feels like Opus 5.5 for coding, while Argon trails on Terminal-Bench and FrontierSWE.

The author suspects Google is using hype around Gemini 4 Argon and the rumored Opus 5.5-class Gemini 4 Carbon to stay in the conversation while delaying access.

A user sarcastically predicts Gemini 4 Argon will arrive in 2027 and Gemini 5 Carbon in October 2127.

A leak says Google is internally testing another model called Carbon after Gemini 4 Argon, and Business Insider reports its coding ability is close to Claude Opus 5.5.

An opinion post argues the AI race is now about availability: Google's Argon is barely accessible, OpenAI's Ultrafast sits behind a $500 plan, and Anthropic is cutting Haiku 5.5 prices to win developers.

A post with a chart says the latest benchmark shows Gemini 4 Argon leading the rankings, but the post gives no specific tests or scores.

A leak claims Google employees are testing a newer Gemini 4 checkpoint called Carbon, with testers saying its coding feels like Opus 5.5, which if true could mean Gemini handles both multimodal and coding work.

The leaderboard shows Gemini 4 Argon leading at 77.9% ahead of GPT-6.1 Sol and Opus 5.5, while Grok 4.7 misses the top three but offers the best value for its price.