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

Reddy shares rumored Gemini 4 Pro pricing of $3 per million input tokens and $12 per million output tokens, arguing an Astra-level model at five times lower cost would shake the market.

An Arabic-language post says that if Gemini 4's test results hold and it really beats GPT-6 Astra and Fable 5, it would be the greatest comeback in AI history and could silence Google's doubters for good.

A meme poking fun at where each model maker stands after Gemini 4 benchmark scores leaked, reflecting community reaction to the leak.

Given the same prompt to build a 3D dragon in three.js on a white background, Gemini 4 Pro keeps pace with Opus 5 at max effort, which the author finds impressive.

A model listed on Arena as gemini-3.7-flash, said to be Gemini 4 Pro, built a 3D Wall-E in three.js and added small animations that were never prompted.

A video meme depicts Gemini 4 after leaked benchmark scores surfaced, playing on reports that its numbers beat rivals.

Rumored Gemini 4 Pro specs include $2.25 per million input tokens and $11.25 output, a 2M-token context, and benchmark wins over GPT-6 Astra and Claude Fable 5.1, none confirmed by Google.

A user shares a screenshot of a new Gemini 4 Pro benchmark, joking that the numbers are too big to display properly.

The same prompt was run on Gemini 4 and GPT-6, with a side-by-side video of the results; the author finds Gemini 4's build very strong.

The author argues Gemini 4 doesn't need to be far smarter, just close to Astra/Fable 5.1 ($10/$50 per 1M tokens) and cheaper, to disrupt frontier model pricing.

A blogger says a suspected Gemini 4 Pro checkpoint is on Arena and collects the strongest outputs seen so far, concluding Google is back.

A poster shares an image with the caption "Gemini 4 Pro?", hinting at another screenshot of output or an interface attributed to Gemini 4 Pro.