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

A video reviews pricing and benchmarks for Meta's new Muse Wonder Full and Muse Wonder models, comparing them with Fable 5.5, Astra 6.1 and Gemini 4 Argon.

A poster says Meta's new Muse Wonder model beats GPT-6 Astra, comes close to Opus 5.5, and also outpaces Gemini 4 Argon.

Meta's Muse Wonder enters the Artificial Analysis leaderboard at 56 in second place, while Gemini 4 Argon sits at 53, behind Opus 5.5 at 58.

A user jokes that Gemini 4 Argon took so long to ship that Meta has already released something better, so Google should skip Gemini 4 and launch 4.1 instead.

A reply says Meta's just-released model is so strong that Google's Argon, reportedly Gemini 4 Pro, has no reason to launch at all.

Claude Opus 5.5 leads the AI Index at 58, followed by Muse Wonder Full at 56, while Gemini 4 Argon, GPT-6 Astra and Claude Fable 5.1 all sit at 53 and Meta jumps from 48 to 56.

Gemini 4 Argon scores 42 on AA-Omniscience and, in its high configuration, has a 15% hallucination rate, the lowest among high-scoring models, per Artificial Analysis.

A user notes that Meta's Muse already beats Opus 5.5 on several benchmarks while Gemini 4 Argon is still unreleased, and questions how long Google expects people to wait.

A Spanish-language post says Meta's new frontier model family Muse Wander is slightly above Claude Opus 5.5 and GPT-6 Astra on some tests and comparable to Gemini 4 Argon, and does not appear to be open-weights.

A reply asserts that Gemini 4 Argon has already lost its competitive edge before release, reflecting community frustration over its delay.

A user jokes that Meta's new model launched before Google's Gemini 4 Argon, with an image that appears to show Meta's release.

A leaked Humanity's Last Exam scorecard shows Meta muse wonder at just 53.6%, Gemini 4 Argon at 57.1%, and Claude Opus 5.5 leading at 61.4%.