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

The author used Gemini 4 Carbon, listed on Arena as gemini-4-argon, to generate 3D code for a mechanical Indian peacock with gears in every feather, calling Carbon the best 3D model right now.

An Arabic-language post says the two non-video models produced a photoreal cinematic shot of a white tiger in a mansion from a single text prompt.

A developer reports Antigravity now offers Gemini 4 Argon plus Claude Opus 5.5 and Sonnet 5.5 in a single IDE.

A post claims Pichai admitted Gemini 3.5 Pro was shelved for weak coding-agent ability, with Google instead iterating 3.6–3.8 Flash fast and feeding that knowledge into Gemini 4 Argon training.

SBI's Chief AI Officer listing asks for 15+ years of experience, including 10+ years hands-on with Opus 5.5, Astra 6.1 and Gemini Argon.

A developer complains GPT's jargon-heavy output causes brain fog, then switches to Gemini 4 Argon for code explanations and finds it plain-spoken and genuinely helpful, planning to route all explanations through a Gemini sub-agent.

An investor's monthly review says Gemini 4 Argon matches GPT-6 Astra and Claude Fable 5.1 but has been overtaken by Meta, with coding hallucinations and tight Pro quotas.

A post asks readers to keep only one subscription among Gemini 4, Claude Opus 5.5, GPT-6 Astra and Grok.

Google Japan's official account says it built a moon-phase simulator with Gemini 4 Argon for the Jugoya-adjacent Thirteen Night moon viewing.

A user says the latest Gemini 4 Carbon checkpoint appears in Arena as gemini-4-argon and shares a Minecraft clone it generated.

A short meme video jokes about Gemini Argon watching a multi-model 'council' from the sidelines.

A comparison post says Anthropic and OpenAI keep shipping, Google returned with Argon and Chinese models are launching, yet there is still no Grok 4.8 news.