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

JakeABoggs says Gemini 4 Argon scored 69% on the hieroglyph benchmark HieroglyphBench, the first model accurate enough to be useful.

A user asked Gemini 4 Argon on the web to make a GIF, and the result came out unexpectedly cute.

The post says Argon is a solid improvement for Google but nowhere near Claude Opus 5.5, and expresses disappointment that Google is still this far behind.

A user testing Gemini 4 Argon in Google Studio says its friendly, non-defensive personality makes it one of the most human and enjoyable models to talk to, while noting Studio's weaker guardrails.

A joking shout-out to Gemini 4 Argon with no actual testing, just a meme-style expression of expectation.

Users are posting screenshots suggesting Gemini AI Pro subscribers can already access Argon, reportedly the codename for Gemini 4 Pro, in the Gemini app, possibly as a limited rollout.

A Spanish-language breaking-news post claims Google has launched Gemini 4 Argon, calling it its most advanced model yet and highlighting coding, cybersecurity and finance tasks.

Third-party platform Dot says Gemini 4 Argon is now live with fully anonymous access, priced 60% below market rates.

PatrickToulme shared thdxr's invitation to try Argon, saying developers' real feedback after using it is better signal on model quality than any benchmark.

The same 5 prompts run through 4 frontier models: Gemini 4 loses the 3D castle and playable game to Opus 5.5, the homepage design to GPT-6 Astra and the short film to Meta Muse Wonder, winning only on writing.

A rundown of Argon access: the paid Gemini API tier, Antigravity for Ultra subscribers, and the Gemini app and Search for Pro and Ultra, with Google citing a 35% rise in agent-written internal code.

A user shares the official Google blog link announcing that Gemini 4 Argon is now live, saying the wait is over.