Gemini 2.5 Pro
Google released Gemini 2.5, which it calls thinking models that reason through their thoughts before responding; the first release, 2.5 Pro Experimental, debuted at number one on the LMArena human-preference leaderboard by a significant margin, and Google said it would build thinking into all its models going forward.
Why it matters
Deliberation before answering stopped being a separate product and became how an ordinary model works.
In half a year from o1, reasoning went from an expensive separate model to a property of the main line at all three large laboratories. The model kept a context window of a million tokens and led on coding, mathematics and the sciences. General availability came on 17 June. For the timeline this is the point where three companies converged on one architectural answer, reached from different directions.