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

A post notes Gemini 4 Argon mini is available at half price and asks who is switching to it.

A thread describes Gemini 4 Argon as built for complex reasoning, coding and multi-step work, with 8 prompts for end-to-end workflows.

With a single prompt, DeepSeek-V4.1-Flash-1017 built a Three.js 3D skeleton tourbillon watch from one image in half an hour, with working gears and refraction, matching Gemini 4 on 3D code.

A developer found computer-use music production in Logic Pro too expensive versus SUNO, and after Gemini 2.5's audio fell short plans to test the newer Argon.

A streamer had Gemini 4 Argon plan his robotics project and hand coding to Opus 5.5 medium without his involvement, arguing Argon's orchestrator role is underdiscussed.

The author tested frontier models on joke writing: Gemini Argon tried too hard, ChatGPT Astra was decent, and Claude Opus 5.5 spent ten minutes without good criteria.

A Japanese user says Gemini 4 Argon successfully resolved bugs and is usable, and expects the situation to shift once Gemini 4 Flash arrives.

A reply guesses Gemini 4 Flash could drop tomorrow, reasoning that Argon was announced on the 30th and three weeks later lands on October 20–21.

An argument that if Gemini Argon matches Claude Opus output at lower cost, Opus faces a real API pricing problem.

Google India's official account posted an image captioned "Gemini Argon gets it," a light official tease for Gemini 4 Argon.

A gemini-4-flash-preview string tied to the Gemma 4 tokenizer has surfaced in Google's SDK, and with Gemini 3.8 Flash promo pricing ending Dec 31, a Gemini 4 Flash launch looks near.

The author says Argon answers obscure esoteric and biological details in one 3-second pass without search, handles coding tasks Opus 5.5 struggles with, and orchestrates Claude and GPT subagents well.