Google launches a preview of Gemini 2.5 Flash in the Gemini app. What are the updates?

Google Gemini 2.5 Flash, its innovative hybrid reasoning model, is now accessible in preview for the standalone Gemini app. Recently, the company released an “early version” of the model featuring enhanced reasoning capabilities. It is designed to more effectively gauge the amount of processing power or “thinking” needed for each request. The Gemini 2.5 Flash model is also available on Google AI Studio and Vertex AI, providing developers with controls to manually disable its reasoning compute and allocate the amount of processing applied to tasks, allowing for greater budget management regarding token usage. Gemini 2.5 Flash is part of the Gemini 2.5 family of models launched earlier this month, which also includes Gemini 2.5 Pro. In the Google model naming convention, Flash is recognized as the smaller, more cost-efficient version of Pro, optimal for daily tasks. Both models in the Gemini 2.5 family feature integrated reasoning skills, automatically adjusted according to the task at hand. The days when Google’s models lagged behind OpenAI’s have long passed (when Gemini was still known as Bard). Currently, Gemini 2.5 Pro leads the LMArena AI leaderboard, with Gemini Flash closely following.

Meanwhile, OpenAI is actively advancing its offerings. Recently, it introduced o3 and o4-mini, the newest iterations of its reasoning models, capable of utilizing all the tools within ChatGPT, including web searches, image recognition, and Python compatibility. It has become typical for both companies to release similar products within a span of 24 to 48 hours, intensifying their competitive dynamic. Moreover, the GPT-4.5 Preview, its latest model that doesn’t utilize chain-of-thought reasoning, holds the second position on the LMArena leaderboard. With Gemini 2.5 Flash now available for testing in preview mode, users are encouraged to explore how it compares to OpenAI’s o-series models.

Topics
Artificial Intelligence
Google Gemini