Google has launched a preview version of Gemini 2.5 Flash, its newest hybrid reasoning model, which is now available through the standalone Gemini application.

Revealed on Thursday, this initial variant of Gemini 2.5 Flash features improved reasoning skills and the capability to adjust the amount of computational power—or “thinking”—it utilizes for each task dynamically. Developers can access the model through Google AI Studio and Vertex AI, where they can choose to manually disable reasoning compute and control the processing power allocated to specific tasks. This offers enhanced control over token usage and overall effectiveness.

Gemini 2.5 Flash is a component of the larger Gemini 2.5 model family, which was launched earlier this month and also features Gemini 2.5 Pro. Within Google’s naming framework, “Flash” indicates a more compact, budget-friendly version of “Pro,” optimized for regular applications. Even though it has a smaller footprint, Flash still incorporates integrated reasoning capabilities that engage automatically based on the task.

This represents a notable advancement from previous versions of Google’s AI models, which faced criticism for trailing behind rivals like OpenAI. Currently, Gemini 2.5 Pro tops the LMArena AI leaderboard, with Gemini 2.5 Flash closely following.

At the same time, OpenAI is also making strides. Just one day prior to Google’s announcement, OpenAI launched its new o3 and o4-mini reasoning models. These models can utilize all tools offered in ChatGPT, including web browsing, image analysis, and Python support. The rapid releases from both firms—often within 24 to 48 hours of each other—underscore the rising competition between the two technology powerhouses. OpenAI’s GPT-4.5 Preview, its final model without chain-of-thought reasoning, currently occupies second place on the LMArena leaderboard.

With Gemini 2.5 Flash now in preview, users can start testing it and evaluating its performance in comparison to OpenAI’s latest o-series models.