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

A user who says they know nothing about coding claims Gemini 4 Argon produced three working local web apps in 30 minutes and shares the localhost addresses.

Given the same puzzle room prompt, Argon's build looks like a tech demo while Opus 5.5 adds textured floors, a detailed portal gun and warm lighting, and the author sees a clear gap.

The post pushes back on criticism of Gemini 4 Argon, saying critics are just Gemini haters and that it performs far better than every Chinese model and even GPT-6.1 Sol.

The author predicts Claude Fable 5.5, GPT-6 Bel and Gemini 4 Argon will top the rankings in a week, and urges Google to release Argon before Haiku 5.5 overtakes it.

The author compares the $20 tiers of OpenAI, Anthropic and Google, complaining that Google only gives access to Gemini 3.8 Flash, implying Gemini 4 has not been released to that tier.

The author posts an Adventure Time-themed video labeled as generated by Gemini 4 Pro, showing its animated-style video generation.

Inmerson says Gemini 4 Pro used just 40% of their weekly quota after completing hundreds of projects, with a screenshot of the quota attached.

If Tibo's tip is accurate, Fable 5.5, Gemini 4 Argon and Astra 6.1 will all arrive next week, though the author still considers Opus 5.5 the strongest.

A user used Gemini 4 Pro to generate a "Kyoto street" video, showing the model's scene and camera generation capabilities.

A leaker says Gemini 4 Argon's release is approaching and may land this week, without offering specific evidence.

The author says this stealth model has been the most-used on OpenCode for five days straight, and in a same-prompt comparison with Gemini 3.8 Flash it was faster and produced a more distinctive landing page.

The author used Gemini 4 Pro to make an interface demo video called Webb OS Part I/II, showing the model's actual performance on UI and motion design.