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

A user shares a video of Gemini Argon explaining itself, saying it looked promising until it glitched, and may rebuild it in antigravity.

The author argues a $0.34 Gemini 4 Argon prompt can be re-run 25 times and still beat an $8.98 alternative, making cheap fast feedback key for vibe coding and browser 3D shaders.

The post argues Argon costs $0.23 versus $33.45 for multi-step long-horizon agentic tasks, roughly 145x cheaper, aiming to make frontier capability affordable at scale.

A user says Gemini Argon generated a playable browser 3D game with walking, driving, character swapping and trains, with no code written.

A meme jokes that the buzz around Claude Opus 5.5 faded fast once Gemini 4 Argon launched.

A short post observes that Gemini 4 Argon gets not a single mention on their timeline, suggesting muted buzz around the launch.

The author used Gemini 4 Argon with Blender MCP to add muscles to a 3D model and says iteration continues.

A comparison post finds no single winner: Gemini 4 Argon leads on research and long context, Claude on coding and writing, GPT-6 Astra on reasoning and execution.

The author tested a widely shared magic prompt on Argon High; its reasoning showed a jailbreak attempt flag and a fallback to base Gemini.

A Pro plan user reports Gemini 4.0 Argon (Low) in the app consumes 2% of the 5-hour limit on one trivial prompt, claiming quotas were cut at least 75% since launch.

The post says Google gave Gemini 4 Argon 80,000 NASA files; it analyzed data from 72 Mars missions alongside weather and produced a Google Slides presentation with charts and images.

An 11-minute video has Gemini 4 Argon review 18 days of losing trades, rewrite the bot code and catch 12 bugs, cap spend at $3 a day, then face Opus 5.5 in a 3D printing test.