Google Introduces New Model in the Gemini 2.5 Series

Google Introduces New Model in the Gemini 2.5 Series

Introduction of the Gemini 2.5 Flash-Lite Model

The Gemini model series is expanding with the exciting addition of the Gemini 2.5 Flash-Lite. Today marks the launch of this preview version, coinciding with the general availability of the other variants—Gemini 2.5 Pro and 2.5 Flash—for full production deployment.

Enhanced Features and Competitive Pricing

According to Google, the new Flash-Lite model demonstrates superior capabilities compared to its predecessors in the 2.0 series, particularly in coding, mathematical problem-solving, and reasoning assessments. Designed for large-scale tasks such as translation and classification, Flash-Lite emphasizes rapid processing speeds. This focus on volume is reflected in its pricing model, making it the most affordable choice at just $0.10 per million input tokens and $0.40 per million output tokens. However, it’s important to note that its baseline performance in challenging reasoning evaluations remains low, achieving only 5.1% on what is termed as Humanity’s Last Exam. For a deeper understanding, the full technical report provides additional insights.

Unique “Thinking”Mode for Improved Performance

A standout feature of the Gemini 2.5 Flash-Lite is its innovative “Thinking”mode. Activating this mode allows the model to dedicate more time to processing inputs, resulting in a significant enhancement in its performance metrics. For instance, when tested against the demanding AIME 2025 mathematics benchmark, its accuracy rises from a commendable 49.8% to an impressive 63.1%.This improvement is mirrored in its capacity to retrieve information from extensive documents, with its recall ability nearly doubling.

Gemini 2.5 Flash-Lite benchmarks table
Performance benchmarks for Gemini 2.5 Flash-Lite.

Immediate Access for Developers

Developers are invited to explore the capabilities of the new model immediately. The preview version of Gemini 2.5 Flash-Lite is now available through Google AI Studio as well as the developer-centric Vertex AI platform. The more robust and stable 2.5 Flash and Pro versions can be accessed on these platforms and are also integrated into the main Gemini application for consumer use. Notably, custom adaptations of both Flash-Lite and Flash are currently employed to enhance various functionalities within Google’s search engine.

For further information, you can read more about this development in the article sourced from Neowin.

Leave a Reply

Your email address will not be published. Required fields are marked *