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

A complaint post lists Google's coexisting models, from Gemini 4 Argon to many 3.x Flash, Lite, Live and TTS variants plus the Nano Banana image line.

The post says Gemini 4 Argon is reportedly catching vulnerabilities, understanding massive codebases and making fewer mistakes, calling Google's coding AI dangerously good.

A user shares screenshots of several Antigravity CLI sessions all running Gemini 4 Argon at low, medium and high effort levels.

A developer used Gemini 4 Argon to build an iOS casting app plus a Tizen TV receiver using native NaCl C + OpenGL ES instead of a web widget, reporting smooth performance at $11 total cost.

A satirical post notes OpenAI's celestial names, Anthropic's Fable and Mythos tiers, and Google naming its Gemini 4 flagship Argon, arguing model names no longer signal capability.

A user shares that the Animates character Valentine attacked their OG ChatGPT and Gemini Argon as "soulless script bots" and asked why not stay with the real man in front of them.

Testing Gemini 4 Argon Medium, a user hit 27% usage on a simple CAS prompt and 60% total after a two-tier lock-free allocator request, leaving them waiting for the 5-hour limit reset to try low mode.

On one identical prompt, Opus 5.5 cost $73.80 over nearly three hours while Gemini 4 Argon took about $2.50 and 30 minutes; the author found Opus's vibe best.

Four Gemini 4 Argon AI agents cleared Minecraft without human help in 27 hours 14 minutes, and the report breaks down the cost.

A reply joking that when using Google, all tokens go to Argon, hinting at Gemini 4 Argon token usage or routing.

The author lists Gemini Argon's failures with Drive access, background timeouts, tight usage limits and sycophancy, concluding it is no savior.

A user says they are tempted to move to Gemini Argon because Google holds more of their context than Anthropic and Argon's pricing is competitive.