Gemini 2.5 Pro: Google builds reasoning into its flagship model
In one sentence Google DeepMind releases Gemini 2.5 Pro Experimental, the first model in the 2.5 family: a 'thinking model' that reasons before answering, with a 1M-token context and a debut at #1 on LMArena.
On March 25, 2025 Google presents Gemini 2.5 Pro, as an experimental release. It is a "thinking model": it reasons before answering, like OpenAI's reasoning models. Google says this capability will be built into all its models from here on.
At launch it debuts at #1 on the LMArena leaderboard, by a margin Google calls significant. It can read up to 1 million tokens at once (2 million announced as coming soon).
It is available immediately in Google AI Studio and in the Gemini app for Gemini Advanced subscribers; Vertex AI availability and pricing are announced for the following weeks.
Companies
Google DeepMind
Tools
Gemini 2.5 Pro
Tags
Sources