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

A video comparison finds Opus 5.5 more realistic and refined, but Argon costs half as much and still looks good, and the author calls it a daily-drivable model that isn't benchmaxed.

Argon is now available on Pro and is a strong model, but one session exhausts the weekly quota and Ultra users hit zero in 1–2 hours, far less durable than Claude and ChatGPT, the poster says.

A post lists how many days each Gemini model from 1.0 to 4 Argon lasted, noting iterations are speeding up: Argon is 3 days old and Carbon is already in internal testing.

After five hours with Gemini 4 Argon, a user lists four complaints: output quality below Opus 5.5, low usage limits, slow speed, and UI/UX needing work, and says they uninstalled Antigravity.

An eight-step tutorial shows how to pick Gemini 4 Argon on sim.ai, deploy it as a chat link, and use the model for free without writing code.

Using the same project via aimlapi.com, the author gives Gemini 4 Argon a physics patch prompt to show an AI physics simulation failing and then being fixed, with the prompt in the pinned comment.

The author says Gemini 4 Argon arrives on the $20 Pro tier today and that they are not judging it on benchmarks but on how long they can use it before hitting the quota limit.

The post says Gemini 4 Argon is strong but too expensive and slow, burning a Pro user's weekly quota on one task, and expects a cheaper, faster Gemini 4 Flash next.

A rumor says Qwen 4 will use RSI across its entire lineup, after Alibaba reportedly tested it on Qwen 3.8 Max for over a month and raised its Intelligence Index from 40 to 45.

A test says Gemini 4 Argon hits the same pixel art problems as Opus 5.5: low detail or nonsensical patterns, flat regions, poor animation, and high cost with long generation times.

Given only one reference photo of a box, Gemini 4 Argon produced a full 3D sushi box effect in three.js and shaders with no 3D models.

A post asks readers to choose between Gemini 4 Argon and Claude Opus 5.5, with the replies showing where each camp stands.