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

The author shows a video polished through multiple iterations over 3 hours with Gemini 3.8 Flash and hopes Gemini 4 will drastically shorten such generation and coding tasks.

On the DeepSWE v1.1 software engineering benchmark, Gemini 3.8 Flash scores 73.7% versus 71.0% for Grok 4.7 xHigh, with the lightweight Flash model beating xAI's flagship.

An image jokes about where other models stand after the release of Gemini 4 Pro, with no actual test results.

A paying subscriber complains that the Gemini subscription feels like a waste of money without a latest Pro model, and says releasing Gemini 4 Pro would make it feel worthwhile.

The author posts two images comparing GPT-6 Pro and Gemini 4 Pro results side by side for reference.

A video shows Gemini 4 Pro producing a 3D model of an Airbus H145 helicopter in under 10 minutes, presented as a 3D generation test.

The user says they are waiting for Gemini 4 Pro and want Google to win, but the post offers only expectations, with no tests or leaks.

A user shares a hands-on video of Gemini 4 Pro, saying Fable, GPT-6 Astra and Grok lag noticeably in the comparison, with a rumor.lab link.

A video showing Gemini 4 Pro in action, which the author describes as having "cooked really hard," with no specific benchmarks given.

A post says a Google DeepMind principal engineer is already trying Gemini 4 Pro, while OpenAI may ship GPT-6 Sol/Luna this week; the rest covers AMD, ARM and other stock trades.

The poster says Logan admitted Gemini 4 Pro is a failure before it even launched, which the poster did not expect.

A video puts Gemini 3.8 Flash and Gemini 4 Pro side by side, offering a rare direct head-to-head comparison.