Google has announced the public preview release of updated versions of Gemini 2.5 Flash and 2.5 Flash-Lite, now available through Google AI Studio and Vertex AI. These models are designed for developers and businesses looking for high-performing, efficient AI solutions for their applications. The updates are accessible immediately to users on these platforms and can be invoked using the provided model strings, with the new "-latest" alias simplifying access to future updates.
The updated Gemini 2.5 Flash-Lite features advanced instruction following, concise response generation, and improved multimodal capabilities, including more accurate audio transcription, image understanding, and translation. Both models significantly reduce token usage, with Flash-Lite cutting output tokens by 50% and Flash by 24%, leading to reduced operational costs and faster response times. Gemini 2.5 Flash introduces better agentic tool use, which supports complex, multi-step processes, and demonstrates a 5% performance gain on the SWE-Bench Verified benchmark. Early feedback from industry users, such as the team at Manus, highlights noticeable improvements in efficiency and performance, particularly for long-horizon agentic tasks.
Google, the company behind this release, emphasizes preview availability to gather user feedback and iterate rapidly on their AI offerings. While these preview versions are not yet considered stable releases, they reflect Google’s ongoing development strategy and commitment to delivering state-of-the-art generative AI tools for the global developer and enterprise community.
