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

A video cites Polymarket's 74% odds that no Pro model ships this year, with 3.1 Pro the last and Gemini 4 Carbon expected next.

Google DeepMind unveiled Genie 4, a real-time photorealistic world model steered by natural language, and pairs it with Gemini 4 Argon and Gemini Robotics to simulate Apptronik's Apollo humanoid playing checkers.

Google DeepMind's Genie 4 world model debuts in a demo where Gemini 4 Argon breaks a checkers task into real-time steps for an Apptronik Apollo humanoid with Sharpa hands.

A user asks whether the €5.50/month student Google AI Pro plan with Gemini 4 Argon, paired with Codex, is worth it for AI agent coding.

Given the last 3% of usage quota, Gemini 4 Argon produced four GPU raymarched worlds and a procedural audio synthesizer in one HTML file, plus a midnight note.

Gemini 4 Flash is reportedly reaching ~300 tokens/sec on LMArena, which the author says would be a major edge if it delivers Argon-level intelligence at Flash pricing.

A Superteam USA event with Google DeepMind is 75% full and promises demos of Gemini 4 Argon and Gemini Audio among latest capabilities.

A Turkish thread details Genie 4's new capabilities—synchronized audio generation, real-time character dialogue, better human motion and physics—plus a Gemini 4 Argon-driven Apollo robot checkers simulation.

A user says Gemini 4 Argon fully understood their guidance while writing copy and gives it praise.

Opus 5.5, Sol 6.1 and Gemini 4 Argon all sit on the Pareto frontier, but Opus is more read-efficient while Argon burns more output tokens to reach the same score.

DeepMind demoed Argon breaking a task into steps and feeding text instructions to Genie 4 as it generates the scene live—a robot simulation experiment, not a physical robot demo.

The author feeds an archive of past tweets to Gemini 4 Argon to clean into atomic JSON, then derives topic-selection and writing skills for batch-reproducing their own style.