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

A user says Gemini 4 Argon generated a koi pond project in 25 minutes at a cost of about $8, and rates it far above GPT-6 Astra.

A developer shares several self-built apps running on localhost ports 3000, 8080 and 5000, made with Gemini 4 Argon, and calls the results wild.

Gemini 4 Argon has launched and is now available to Pro and Ultra subscribers directly in the Gemini app.

Applications are open for scientists and researchers, with 10,000 slots offering six months of Google Antigravity's highest tier, including access to Gemini 4 Argon.

A post says Gemini 4 Argon has a free allowance on Experiential, where users can sign in to check eligibility and try it in the playground.

Multiple users complain that Gemini 4 Argon's output quality has dropped and limits have tightened, questioning whether Google quietly downgraded it after release.

Odds on a prediction market for Gemini 4's public release by October 15 rose from 54% on Saturday to 94%, though Google only announced a limited rollout with no public date.

A user complains Argon is priced like Sonnet and Sol and cheaper than 3.1 Pro, yet Google treats it as a flagship locked to the $100 Ultra tier while free users are pushed down to Flash-Lite.

A Japanese creator redid their base prompts, character data, and outfit detail rules on Gemini 4 Argon, using Google Flow for backgrounds, and shared the resulting illustration.

SpeechMap, which tracks which controversial topics labs let models write about, gives Gemini 4 Argon 87.0%, with Google's scores staying more consistent than other labs'.

A writer ranks Gemini 4 Argon against Claude Opus 4.6 and 5.5 across power fantasy, light novels, fan fiction, and strong-emotion scenes, finding Argon nails the opening hook but pushes plot on its own and overdoes character reactions.

A paying Pro subscriber says Google should at least offer limited Gemini 4 Argon requests to Pro users instead of reserving it entirely for Ultra.