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

A post claims Meta's new model Wonder beat GPT-6 Astra and Gemini 4 Argon and trails Claude Opus 5.5 by only two points.

A NYC test of Meta Muse Wonder on max shows glitched visuals, and Artificial Analysis data puts Gemini 4 Argon at $1.99 per task, below GPT-6 Astra and Opus 5.5.

Bind Reddy previews LiveBench 2.0, focused on long-horizon agent tasks and cost per task, saying it will launch the same day as Gemini 4 Argon.

A video puts Meta's new Muse Wonder Full model up against GPT-6 Astra and Gemini 4 Argon for a side-by-side comparison.

A poster says Meta's Muse Wonder stole the spotlight, leaving Gemini 4 Argon over before it began, and sees Meta leading in personal AI.

Muse Wonder has overtaken the still-unreleased Gemini 4 Argon on the Artificial Analysis index, leaving the poster to wonder whether Google ships Argon tomorrow or skips to Gemini 4.1.

After Argon and Carbon, the poster guesses whether the next Gemini 4 releases will be called Neon, Xenon or Krypton, reflecting community discussion of the naming scheme.

A user points out that Mark teased and released muse wonder on the same day, while Google teased Gemini 4 Argon three weeks ago and it still hasn't launched.

The classic pelican-on-a-bicycle prompt is used to test Argon's SVG generation, with the output shown in an attached image.

A user shares a comparison chart joking that the unreleased Gemini 4 Argon is already shown as falling behind.

Hesamation says Meta's new Muse Wonder model ties Sonnet 5.5 on intelligence score and knocks Google out of the top three before Gemini 4 Argon is even released.

A poster shares a prompt that queries the gen_metadata endpoint identifier, where gemini-3.8-flash-exp-a means Argon, but five accounts all missed, suggesting access may be limited to paid Pro users.