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

Google released Gemini 4 Argon to compete with GPT-6 Astra and Claude Opus 5.5, with Ultra access in Antigravity and Pro users now able to try it on web and mobile.

A user says three messages with Gemini 4 Argon consumed 8% of their Pro plan quota and that the model already hallucinated and mixed languages.

A user complains that Gemini 4 Pro Argon is only partially available to Ultra subscribers and says they are more looking forward to Fable 5.5.

After comparing about 20 Japanese articles, the author finds Opus 5.5 more accurate and cheaper, while Gemini 4 Argon reads well but tends to rewrite by guesswork and hallucinate.

A post containing only the text "Gemini 4 Argon" and an image, with no explanation, so the image has to be opened to see what it shows.

A user found that a few-yuan Gemini Pro account on Xianyu could access Gemini 4 Argon, and within hours sellers had raised prices several times over, up to 48 yuan.

A video comparison shows Opus 5.5's gummy bears stretching, bouncing and looking translucent like jelly, while Gemini 4 Argon's are stiff and clip through each other.

Ethan Mollick asked AIs for their best "unless it's from X, it's just sparkling Y" jokes, with Gemini 4 Argon High's answer lined up alongside GPT-6 Pro and Opus 5.5 Max.

In a demo, Gemini 4 Argon watched a human play cornhole, built its own physics simulation and generated control logic, letting an Apollo humanoid robot land 7 consecutive throws.

freebuff added Gemini 4 Argon to its coding agent, offering one hour of daily use for tier 1 countries and claiming the model maxes out most benchmarks.

A user made an ASCII animation of a seed sprouting into a tree with Gemini 4 Argon, remarking that the model may not be benchmark-tuned after all.

Google Cloud's official account said Gemini 4 Argon is now available on Google Cloud and had Google Developer Experts show how they use it to transform developer workflows.