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

A Japanese user says Gemini 4 Argon is officially out: its benchmark scores only catch up with top models, but its very large context window gives it an edge on long documents and repeated reasoning.

A user says Gemini 4 Argon feels so different that it raises the question of how many models are just copying the work of Opus 5.5 and Sol 6.1.

A short hands-on post in which the author says they are starting to test Gemini 4 Argon today, with no conclusions yet.

A post lists v0.app and platform.xplabs.ai as two platforms where Gemini 4 Argon can be used for free to write code.

A Japanese user compared Gemini 4 Argon, Claude Opus 5.5 and GPT-6.1 on the same prompt, concluding it isn't worth switching from Claude Code and Codex, though it would be a main tool for game making and video editing.

A user used Gemini 4 Argon to generate an abstract self-portrait titled "87.3K liquid argon Cherenkov wake and Lorentz spirals," as a creative image-generation test.

A user says the Gemini 4 Argon model itself is impressive, but the Gemini app's safety framework is overly defensive, triggering lecturing disclaimers or refusals even on AI internal states and open philosophical questions, which the user calls epistemic gaslighting.

The author used Three.js and WebGL to have Gemini 4 Argon, Muse, Claude Opus 5.5 and GPT-6 Astra simulate the margarita sloshing effect on the same task, saying the gaps were clear and Gemini has improved a lot.

XPLabs offers about 500K free output tokens per day for Gemini 4 Argon, but a card and $1 verification are required, after which standard API pricing applies.

The author made a smartwatch ad with Gemini 4 Argon and Canvas, relying on Canvas's built-in TTS and a Nanobanana image proxy, and generating only the music in a separate chat before combining.

The author says their own tests put Gemini 4 Argon behind Astra for frontend work, ranking Opus 5.5 above Astra above Argon, contrary to online claims of impressive results.

Given the same prompt for a soft-body collision simulation, Opus 5.5 rendered deformation, rebound and light transmission, while Gemini 4 Argon clipped and looked rigid, at $22 versus $7.2.