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

A user with work access to Gemini Argon Pro says it is powerful but has no middle ground, producing either huge analyses or tweet-length answers.

A long breakdown of Genie 4: live editable worlds, native audio, more realistic humans, with Gemini 4 Argon decomposing tasks and 3.8 Flash directing, still research preview at ~2 minutes of consistency.

A developer lists what one Gemini API key enables for agents—image/video understanding, audio, generation, TTS, cheap transcription and classification—and says Argon 4 is creative and uncorrelated with Claude/OpenAI models, near frontier but still behind Opus 5.5 for daily use.

DeepMind's real-time world model Genie 4 adds native audio, with Gemini 3.8 Flash directing the loop and Gemini 4 Argon simulating a humanoid robot playing checkers.

Commentary argues that if Gemini 4 Flash delivers a big speed jump without hurting reasoning and coding, the upgrade could be larger than expected.

On a 100-task benchmark spanning 55 microscope slides and 7 tumor types, Gemini 4 Argon ranks first with 32% pass@1.

A brief reaction claiming Gemini 4 Flash is genuinely fast, without concrete numbers or test details.

Rumors say Gemini 4 Flash is testing internally as Hydrogen, with a gemini-3.8-flash-router model ID and speculation of a launch this week or next.

A leak says Gemini 4 Flash has appeared internally under the codename Hydrogen, with no sign of Gemini 4 Carbon and speculation that the Arena Argon checkpoint was actually Flash.

A user posts screenshots of trying Gemini 4 Argon after signing up through a student offer.

Google confirmed Gemini 4 Argon is gradually rolling out to Antigravity IDE users, with the update expected within 24 hours.

The author says Gemini confidently gave hours and an address for a nonexistent coffee shop without searching, so their bar for Gemini 4 Flash is simply to look things up before answering.