Google introduces Gemini 2.5 Pro, the first reasoning model in the 2.5 line
Gemini 2.5 Pro Experimental reasons before answering, offers a one-million-token context window and leads several maths, science and coding benchmarks.
Google has introduced Gemini 2.5 Pro, described by the company as the first reasoning model in the 2.5 line. This version, named Experimental, led several benchmarks in maths, science and coding at the time of launch.
Two points stand out. First, the model spends time thinking before answering, which suits multi-step problems better than quick lookups. Second is the one-million-token context window, meaning the ability to take in a very large amount of text in a single request. For enterprises, this opens up ways of working with long documents such as project dossiers, contracts or internal procedures without cutting them into disconnected chunks.
The practical significance lies in tasks that require cross-referencing: checking a long contract against its annexes, finding contradictions across many documents, or question answering over a technical document base. When the model holds more context, engineering teams spend less effort on pre-processing and on stitching results together from multiple calls.
Large context does not guarantee accuracy, however, and benchmark results do not replace testing on real data. Costs are usually billed by token volume, so long documents also raise the cost of each call. Sending internal documents to a cloud service needs to sit within a clear security policy.
Source: blog.google
