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

The creator recreated a Seedance 2.5 baby-dragon fantasy video using Gemini Argon and posted it as part one of a series.

A meme contrasts OpenAI and Anthropic naming schemes with Google's Gemini 4 Argon and Carbon, joking it names models after elements.

A user posts screenshots of trying Gemini 4 Argon after signing up through a student offer.

An image post complains that Gemini 4 Argon is getting dumber day by day, without offering specific tests or data.

A user spotted a new gemini-4-argon-c routing model in Antigravity, suspected to be the upcoming Gemini 4 Carbon, with a visible pelican-on-bicycle improvement.

The author argues OpenAI has ceded high-end enterprise competition, positioning GPT-6.1 Terra to defend budget users against Gemini 4 Flash and DeepSeek V4.1 Flash.

A video post claims Gemini 4 Argon casually beats GPT-6 Astra on a new computer-science research benchmark, with comparison footage attached.

The same supercar prompt run through each lab's own agent: Gemini 4 Argon took 8m41s and $0.48, succeeding on the second try.

A long post details Gemini 4 Argon's 1M-token context and output limit, the Gemini Spark background assistant, four free access routes (AI Studio free tier, student plan, family group, Cloud credits), and tips to conserve quota.

A user finds Pixel-account Gemini Pro and Antigravity Argon calls actually hit gemini-4-argon-c, guessing it is Carbon, with a pelican test image attached.

A user post says Gemini 4 Flash's internal codename is Hydrogen, hinting at an imminent launch.

Google confirmed Gemini 4 Argon is gradually rolling out to Antigravity IDE users, with the update expected within 24 hours.