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

A Turkish user reports Gemini 4 Argon at medium effort solved two FPS bugs in 1 hour 20 minutes that GPT-6.1 Sol (high) had failed to fix for six days.

A question post asks readers to pick the best model for complex coding among GPT-6 Astra, Claude Fable 5.1, Gemini 4 Argon and Grok 4.7.

A short post asking how Gemini 4 Argon is and whether anyone has tried it yet.

A short post noting that Gemini 4 Argon's buzz died down very quickly and asking what happened, reflecting community disappointment.

A 5x Ultra subscriber says Gemini 4 Argon quotas in Gemini and Antigravity are too low for large agentic experiments, and model availability also needs fixing.

A footnote in Google's Antigravity docs says Gemini 4 Argon is unavailable in the standalone IDE, directing paid Pro users to the desktop app, agy CLI or VS Code extension.

On EdisonSci's long-horizon bioinformatics benchmark BixBench3, Gemini 4 Argon took the top spot this week as recent releases pushed the Pareto frontier.

On the same prompt, Opus 5.5 produced a jelly ghost while Gemini 4 Argon made a plastic bag with eyes, costing $2.7 versus $0.32.

A developer used Antigravity, Gemini 4 Argon, and img2threejs techniques to build a playable Roadster night-drive scene on the web.

The post lists Antigravity quota tiers: Argon Ultra max20x around $800 weekly, AI Pro $10 per 5 hours and $40 weekly, AI Ultra 5X at $50/$200.

A user spending thousands monthly on AI ranks models from best to worst, calling Opus 5.5 all-round and GPT Astra poor at UI, while noting he hasn't tried Muse Wonder or Gemini Argon.

The author argues that after months of Argon 4 hype, complaints about usage limits, guardrails and coding experience could send Gemini users to Claude.