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

A critic says Google announced Gemini 4 Argon 11 days ago without releasing it, while Meta's Muse Wonder is usable at launch and scores better on paper.

A user complains that Muse Wonder Full is benchmaxxed and says they still hold out hope for the unreleased Gemini 4 Argon.

A reply says the author has a feeling Gemini Argon will do far better in practice than the results currently going around.

A video meme jokes that Google engineers plan to abandon Gemini 4 Argon after seeing Muse Wonder.

Meta's new Muse Wonder scores 56 on Artificial Analysis, tying Sonnet 5.5 and beating Gemini 4 and GPT-6 Astra by 3 points, though the poster suspects it is just a benchmark model.

A poster says Muse Wonder beats Opus 5.5 and GPT-6 Astra on several benchmarks and matches Gemini 4 Argon, but unlike Google, Meta actually shipped the model.

The author, who uses OpenAI, Anthropic and Google daily, says Gemini isn't yet as intelligent as the first two but Gemini 4/Argon might become a true rival.

The post says Muse Wonder and Wonder Full both use fewer tokens than Gemini 4 Argon while scoring higher on intelligence, calling it a breakthrough for Meta.

The poster says Meta teased Muse Wonder and released it hours later, beating Gemini 4 Argon on all benchmarks, while Argon is rumored to launch publicly tomorrow.

A user asks which companies are testing Argon and why they aren't using stronger models to test with.

The user argues that Argon, reportedly Gemini 4 Pro, is already behind Meta's Muse Wonder on paper and suggests Google skip it.

Meta's Muse Wonder scores 56 to 53 over Gemini 4 Argon on the AA Index, but early tests show weak 3D and frontend output, suggesting benchmark gaming.