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

The account RealFedeURU posted a video captioned only "Gemini 4 Argon," apparently a demo or leaked footage of the model.

A list of models predicted to launch soon includes Gemini 4 Argon (possibly a public release), GPT-6.1 Astra, Claude Fable 5.5 and others.

A Korean-language post questions why Gemini 4 Pro, previously hyped as Astra-level and beating every frontier model, only narrowly outperforms Opus 4.8.

A user spotted Gemini 4 Argon in the Gemini API documentation and speculates it could launch next week alongside Fable 5.5 and GPT-6.1 Astra.

A user shares a Gemini 4 benchmark screenshot, saying Google can skip incremental releases like it did last year, but expects the next OpenAI or Anthropic model to close the gap or pull ahead.

An October 4 AI news roundup says Google's Gemini 4 Argon is currently open only to cyber defenders, alongside items on an Anthropic IPO, OpenAI agents and more.

The author argues Argon disproves the claim that Google left the frontier, scoring 53 to tie GPT-6 Astra and trail Opus 5.5's 58, strong on long documents and legal work but weak on terminal coding and frontend, with in-house TPUs keeping its $2/$10 pricing below rivals.

A question post asks why many people in tech have recently been switching from Claude to Gemini 4 Argon.

A post claims that Gemini 4 Argon, Google's very limited frontier model, is expected to open to the public next week or very soon.

A US market close recap mentions that Google's Gemini 4 Argon, released this week, took top benchmark scores, with a cautious phased rollout starting with cybersecurity partners.

An Arabic post says that from October 9 free users will lose access to Gemini 3.8 Flash and 3.1 Pro, keeping only Flash-Lite, while AI Plus also loses Pro and Deep Think goes to AI Pro.

A reply says Google should keep only Gemini 3.8 Flash, calling the other Flash models and even the Pro one outdated, and release Argon right away.