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

A user gave Gemini 4 Argon another complex task inside Antigravity to show how it performs in practice.

A user posts a meme and asks whether the old joke about Gemini still holds up now that Gemini 4 Argon is out.

A Chinese post lists Gemini 4 Argon among the well-known models outmatched by DeepSeek V4.1 Flash 1017 on combined performance and price, noting the latter's near-lowest pricing.

The author argues 2,783 self-written programs measure output rather than reliability, and wants failed runs disclosed before judging Gemini 4 Argon's coding.

A developer is asking for donated Gemini 4 Argon API credits to test the model's robot capabilities in real time.

Users say AI Pro obtained via Pixel purchase, Jio bundle or student offer cannot use Gemini 4 Argon in Antigravity, and the same account cannot switch to a regular paid Pro plan until the offer expires.

A user tests Gemini 4 Argon High by generating an SVG of ten stacked riders, from a pelican on a bicycle up through a scooter, flamingo, tricycle, turtle and more.

The author argues stronger models reward clear work standards over long prompts, listing ten briefing items and naming Gemini 4 Argon among recently advanced models.

A video post claims a Jev-orchestrated stack uses Gemini 4 Argon to ingest 10M-token corpora and figures alongside Opus 5.5 and GPT-6.1 Sol, citing token savings and accuracy figures.

A Japanese tech outlet reports that Google has begun general availability of its top-tier Gemini 4 Argon model.

A developer choosing an AI to build a minigame is torn between Gemini 4 Argon, Claude Opus and Codex, saying Argon looks solid so far.

Google split Antigravity quotas so Gemini 4 Argon runs on its own weekly-only bucket separate from 3.8 Flash, and long agentic sessions burn through it fast.