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

A short image post calls Gemini 4 Argon the first AI model with zero chill, poking fun at its unrestrained output style.

A user bought a Taiwan-region Google AI Pro plan to test Gemini 4 Argon limits; one deep-thinking pelican-on-bike SVG task consumed about 100k tokens, roughly 3% of the weekly quota.

The author ran five new models, including Gemini 4 Argon, as parallel research agents: all results arrived in 21 seconds versus 80 serially, with 5/5 first-pass format compliance, but the slowest task capped the speedup at 3.8x.

With one prompt, Gemini 4 Argon rendered an ornate rhino as a glowing armored rose while Opus 5.5 produced a restrained realistic one-horned rhino.

The poster says Google's favorability beat expectations but questions the methodology, arguing Gemini 4 Argon usage alone would make it a massive success.

A leaker says Google's Gemini 4 Carbon is nearing launch as its real answer to Opus 5.5, possibly within weeks.

A user says just a week after Gemini 4 Argon, Google is expected to ship Gemini 4 Carbon within weeks, with partners already given early access for testing.

A post claims Gemini 4 Argon processed roughly 25 trillion tokens in its first 72 hours, with the author declaring that Google is back.

A leak says Gemini 4 Carbon is now in LmArena, partners have early access, and wider availability should follow soon.

A user measured the Antigravity Ultra 5x weekly quota at about $310, with Gemini 4 Argon consuming $185.88 and Gemini 3.8 Flash $124.61, concluding Argon drains quota fast and Ultra 20X is better value.

A short post notes that Gemini 4 Argon has barely launched and Google is already preparing its successor, codenamed Carbon.

The author converts each subscription to API-equivalent value at max reasoning: Claude Max 5x ~$5,725, ChatGPT Pro ~$1,055, Google AI Ultra (Argon) ~$300, then derives per-task cost.