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

The author had Gemini 4 Argon and Opus 5.5 each build a house and a llama in Minecraft; the API cost was $3.63 for Argon and $7.25 for Opus, and they found Argon had more attention to detail.

The author used Gemini 4 Argon to generate a Minecraft mod that lets you play as any version of Spider-Man in game, calling the result impressive.

Given the same prompt to build a 3D boss fight of a small knight against a clockwork colossus in the sky, Gemini 4 Argon and Muse Wonder produced results described as a studio trailer versus a prototype.

Workflow tips for using Argon in Antigravity CLI: adjusting reasoning effort, one-shot model consults, delegating to subagents, /btw side questions, and /diff change review.

The author tried player re-identification with Gemini 4 Argon, starting by matching the same player across two images, and shared a video demo.

A user shared images generated with Gemini 4 Argon and said it performs quite well at creating art.

The post says Gemini 4 Argon is priced like Opus 5.5 but outperformed by Sonnet 5.5 and Sol 6.1 and behind open models like Kimi K3 and Qwen 3.8 Max, calling it one of the most disappointing releases.

A user says Gemini 4 Argon outperforms Opus 5.5 in their non-coding work, but complains that Google makes it hard to get started and that usage costs can take over 24 hours to appear.

A user is collecting 3D examples made with Gemini 4 Argon, saying they have seen many cool ones and will round up favorites in the thread.

The author shares an image generated with Gemini 4 Argon titled "THE EMPYREAN OVER-LOOM."

Sergey Brin speaking publicly about Gemini 4 Argon prompts discussion about whether Google is preparing a major push.

A user shared screenshots from testing Gemini 4 Argon and said the experience feels closer to AGI than anything before it.