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

A user says a new Gemini 4 Pro checkpoint showed up just days after the previous one, and after trying for hours they still haven't hit it but will share outputs once they do.

A new Gemini 4 Pro checkpoint has surfaced on Arena under the name gemini-3.8-flash, the third such placeholder model in recent days.

A new Gemini 4 Pro checkpoint has appeared on Arena, with leaks saying it performs worse than the previous one; the poster plans to test it and share results.

A video jokes that people are still anxiously waiting for Gemini 4 Pro, reflecting the community's anticipation and impatience.

The user criticizes Google for replacing large Ultra models with medium-sized Pro ones and says every release is nerfed on day one, so the new checkpoint brings no real improvement.

The poster says too many AI models are launching back to back and notes that Gemini 4 Pro is apparently next.

An image jokes about the industry reaction after Gemini 4 Pro's launch, hinting at awkward times for other model makers.

A post says that if the leak is accurate, Gemini 4 Pro outperforms GPT-6 Astra and Opus 5.5 while costing a quarter as much, which could let Google overtake the field overnight.

An image post jokes about how people reacted once Gemini 4 Pro went live, a snapshot of community mood on launch day with no benchmark details.

The author lists four flagship models due for an update, including Gemini 4 Pro, and asks which one will be succeeded first.

An image post jokes about people waiting for Gemini 4 Pro, with the caption saying the real fun only starts after it arrives.

Rumors point to a mid-to-late October launch for Gemini 4 with xhigh and max reasoning modes, a 1M-token context window and 256K output, while the barium-b checkpoint has reportedly been pulled from Arena.