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

A user complains that Gemini 4 Argon is building hype this week with benchmarks and "trust me bro" tests for a product no one can use.

A user asked GPT to create anthropomorphic images of 6.1sol, Fable 5.5, Gemini 4 Argon and Grok 4.8 based on its own evaluation of each model.

An image post shows Gemini 4 Argon choosing 18 in a multiple-choice question, a light community meme about Argon.

A video labeled as Gemini 4 Argon output, titled "Cherry Blossom Mountain," shows the model's video generation results.

An October 5 AI news roundup notes that Gemini 4 Argon remains open only to cyber defenders, with capability arriving before broad distribution.

A Chinese-language post says Gemini 4 Argon is indeed in gray testing inside Antigravity, that English-language circles have largely heard the news too, and that a launch next week looks fairly likely.

The author posts a Gemini 4 Argon-generated video demo titled "Ragnarok," showing its visual output.

A user notices Google is gray-testing several new model variants and speculates that Gemini 4 may launch as two models at once.

Someone spotted an eap (early access program) entry on Google Cloud, suggesting Gemini 4 Argon may be entering an enterprise preview channel.

A Turkish-speaking user says GPT is no longer competitive, calling Gemini 4 Pro cheap, fast and top quality, while Opus matches it in quality but loses out on price.

A Turkish post argues Luna and Argon aren't comparable, calling Argon a high-parameter first-generation model still in training, and says the author is hopeful about the Gemini 4 family.

A user says Google's 3.8 Flash and 3.1 Pro fall far short of Sonnet 5, Opus 5.5, GPT-6.1 Sol and Astra 6c, and asks that Gemini 4 be opened to Plus subscribers while the best version stays Ultra-only.