Google announces Gemini 3.1 Pro: «a step forward in reasoning»

Expect more, rapid one-point releases from the frontier labs going ahead. (Picture: Google)
The new model is a «smarter, more capable baseline for complex problem-solving,» Google says.

It’s the new leader on the benchmarks, almost solving ARC-AGI with 98% and hitting 77% on ARC-AGI 2, the problem solver benchmark, more than doubling Gemini 3’s results.

It also scores 44.4% on Humanity’s Last Exam, which is the best yet and definitly makes it the new state-of-the art.

Gemini 3 Pro was only released last November, so this new point-release might well signal where frontier labs are heading next — incremental updates every few months rather than full blown giant steps every six months to a year.

Gemini 3.1 Pro is rolling out today «in preview» to Pro and Ultra plans, and it’s also in NotebookLM for those users. Developers should also get a taste soon. Pricing is the same as Gemini 3 Pro.

Read more: Google’s announcement, launch thread. See also 9to5Google, Ars Technica and TechCrunch. Discussion on r/Singularity.