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

The author questions why Google locked the newly released Gemini 4 Argon behind the $100 Ultra subscription, leaving $20 AI Pro users able to see it only in the Gemini app.

A developer shares a video of a 3D portfolio website built fully with Gemini 4 Argon, offered as a hands-on reference for its frontend and 3D coding ability.

A Japanese creator made a Strata explainer video with MiniMax-H3 for visuals and Suno for music, with choreography handled by Claude Opus 5.5 and Gemini Argon.

The post says Gemini 4 Argon High ranks first on the Arena text leaderboard with 1525 points, and Polymarket gives Google a 72% chance of having the best model by the end of October, though many commenters are unconvinced.

The poster says Gemini 4 Argon output costs $10 per million words until November 30 and complains that nobody told them about the deal.

The post says Google's Deep Review feature now uses the new Argon model and offers two free paper reviews per week.

Epoch AI says Gemini 4 Argon scores 165 on the ECI, level with Fable 5.1 and behind Opus 5.5 (167) and GPT-6 Astra (166), but a big jump from Gemini 3.8 Flash's 157.

The author gave Gemini 4 Argon and Claude Opus 5.5 the same 30-second video task, framed around which model better understands the vibe, and asks viewers to pick a side.

BridgeBench reports Gemini 4 Argon scored 623 for 8th place, losing to GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5, and ranking below June's Fable 5.

In a clip, Gemini 4 Argon is said to have researched early childhood learning theory and built a knowledge base before writing the frontend and deploying a literacy app, rather than assembling a one-off app.

A Persian-language post shares an Ultra Reasoning prompt refined with Gemini 4 Argon, which the author says cuts model hallucinations and error rates sharply for use in chats or agents.

A Spanish-language reply says something had more impact than Google's fiasco with Gemini 4 Pro, without naming the event.