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

A poster calls on Google DeepMind to release Gemini 4 Argon immediately, saying OpenAI and Anthropic are about to launch Fable 5.5 and 6.1 Sol Ultrafast.

The post argues Argon has strong momentum now, and if another model launches first the attention will fade, so Google should release it as soon as possible.

A post lists three rumored launches for next week, Fable 5.5, GPT-6.1 and Gemini 4 Argon, and asks which will take the lead.

A poster with a video calls on Google DeepMind to make Gemini 4 Argon available to the public, saying they have a few ideas.

The author says Gemini 4 Argon outperforms GPT-6 Astra and Opus 5.5 in every area except agent coding, with aggressive pricing, and that agent coding is exactly where Claude is strongest.

LuminaBench says Gemini 4 Argon is still rolling out to trusted testers, with select accounts seen testing it in Antigravity, and complains about the wait after the hype and benchmarks.

Gemini 4 Argon took first place on the Vals Index at 68.9%, the first Gemini model to reach the top, and also ranked first on Finance Agent v2.

A video says Gemini 4 Argon is releasing publicly soon and that Google deployed Argon agents internally that autonomously identified up to 1 petabyte in memory savings across its data centers.

A Gemini 4 Pro generated video styled after Samurai Jack shows the model's ability to imitate an animation style; the source and prompt are not given.

A Chinese post says that days after Gemini 4 Argon launched, Google cut free users to only 3.5 Flash-Lite, with Argon limited to Fairwind security partners before paid API and AI Ultra access.

A user notes that Gemini 4 Argon can output up to 1M tokens in a single response and asks what task actually needs that much output.

The author checked the HUD readings frame by frame and says the physics numbers in Gemini 4 Pro's binary black hole merger simulation match in both direction and magnitude.