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

On one burning-car prompt, Opus 5.5 took 20 minutes, 1.14M tokens and $3.45, while Gemini 4 Argon took 29 minutes, 2.46M tokens and $2.20, with different visuals.

A leak says Gemini 4 Flash, codenamed Hydrogen, is coming, giving Pro subscribers more generous usage than Argon's daily cap.

After heavy use of Gemini 4 Argon, a user says they regained confidence in Google, citing less distillation and more flair, and is watching the upcoming earnings.

A commentator says Gemini 4 Argon has caught GPT-6 Astra on Artificial Analysis, and if Carbon beats it Google could overtake OpenAI for good.

A roundup of Sonnet 5.5, GPT-6.1 Sol, Gemini 4 Argon, JetBrains Air and Copilot Computer Use, with advice on picking combinations over chasing new models.

A promo post says Freebuff gives free access to multiple models including Gemini 4 Argon with no credit card or top-up, ad-supported with rate limits.

A Turkish-language post asks whether FactoryAI is actually serving users GLM 4.6 behind what it calls Gemini 4 Argon.

A user claims Gemini 4 Argon already matches GPT-6 Astra on AA benchmarks, with a stronger Carbon and a Hydrogen-codenamed Flash coming.

Rumors say Gemini 4 Flash is testing internally as Hydrogen, with a gemini-3.8-flash-router model ID and speculation of a launch this week or next.

Rumors around Gemini 4 Flash grow, with early Carbon testers saying its 3D output beats Argon but runs slower; the author guesses Flash pricing at $0.50–0.75 input and $2–4 output per million tokens.

Ad-supported client freebuff gives free access to several models, including one daily Gemini 4 Argon session, with rate limits.

A user predicts Google will surpass OpenAI by year end with Gemini 4 Argon, migrating all GPT-6.1 Sol, Terra and Sonnet 5.5 workloads, and says the upcoming Gemini 4 Carbon will beat Opus 5.5.