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

In a video, Google co-founder Sergey Brin says the company has been on a recursive self-improvement path since Flash 3.6/3.7/3.8, and that internal models are already several generations ahead of the Argon the public sees, with Logan Kilpatrick agreeing on the 3-4 week release cadence.

A new independent benchmark tests strategy and computer use in Stronghold Crusader, with Gemini 4 Argon at 62.3 ahead of GPT-6.1 Sol's 56.9, and all code and recordings released openly.

A Japanese post says Argon has arrived, requiring a plan of 16,000 yen or more on Antigravity while the browser version can be tried on a roughly 3,000 yen plan, though usage drains fast, with half the 5-hour limit gone after 3-4 tasks.

A user reports that Gemini 4 Argon responds slowly and that the Antigravity CLI shows only a few lines for long stretches, with no visible sign of the model's thinking.

The author gave the same prompt to Gemini 4 Argon, GPT 6.1 Sol and Claude Opus 5.5 to build Google Slides, finding Opus 5.5 strongest overall on design, GPT 6.1 Sol smoother in flow, and Argon strong in places but quirky.

The author used DeepSeek v4.1 Flash to generate a Three.js animation of the ISS Canadarm2 grabbing an approaching cargo craft, rating it below Opus 5.5 and Gemini 4 Argon but around GPT-6.1 Sol's level.

Blazerail says users can claim 2 million free tokens to test Gemini 4 Argon with no credit card required, and the poster plans to try it personally.

A user says Gemini 4 Argon is now available on HIX AI with free, unlimited access for a limited time, lowering the barrier from idea to hands-on testing.

HIX AI announced it has added Gemini 4 Argon with free, unlimited use, and users can compare it with other models in the same workspace.

The author says Gemini 4 Argon is now on HIX AI with free, unlimited access, and plans to test how it performs on real tasks.

Third-party platform HIX AI announced it has added Gemini 4 Argon with free, unlimited access, letting developers run model comparison experiments without paying per call.

Gemini 4 Argon's launch price of $2/$10 per million input/output tokens rises to $4/$20 after November 30, so long-running tools should be budgeted at the regular rate.