Lenovo Flagship Chromebook, 14'' FHD Touchscreen Slim Thin Light Laptop Computer, 8-Core MediaTek Kompanio 520 Processor, 4GB RAM, 64GB eMMC, WiFi 6,Chrome OS, Abyss Blue
50% OffGoogle has announced the rollout of upgraded versions of its Gemini 2.5 Flash and 2.5 Flash-Lite models, which can now be accessed on Google AI Studio and Vertex AI. These enhancements are designed to enhance both the quality of outputs and operational efficiency for users.
The revamped Gemini 2.5 models have made significant strides in performance compared to their predecessors. Gemini 2.5 Flash-Lite boasts a 50% reduction in output tokens, which directly translates to lower operational costs, while the standard Gemini 2.5 Flash sees a 24% cut in token usage.
The latest iteration of Gemini 2.5 Flash-Lite is particularly focused on three primary improvements: it exhibits enhanced adherence to intricate instructions, achieves more succinct responses to minimize token expenditure and latency, and showcases advanced capabilities in multimodal tasks and translation accuracy. Users can try out this version immediately by using the model string: gemini-2.5-flash-lite-preview-09-2025.
Meanwhile, the updates to Gemini 2.5 Flash revolve around two critical areas that have traditionally drawn user feedback. The model now demonstrates better tool usage, which enhances its performance in complex, multi-step applications. Early assessments show a notable improvement, with a 5% increase on the SWE-Bench Verified metric, rising from 48.9% to 54%. Additionally, this version significantly reduces costs by generating higher quality outputs with fewer tokens when “thinking on,” thus minimizing latency.
Early testers like Yichao “Peak” Ji, Co-Founder and Chief Scientist at Manus, have praised the new model, highlighting a 15% performance increase in long-term tasks. Ji emphasized how the cost efficiency of the new model enables Manus to expand its operations and fulfill its objective of enhancing human capabilities.
To facilitate seamless access to new models, Google is also introducing a “-latest” alias for each model family. This feature directs users to the most current versions, allowing them to experiment without needing to constantly update their code with new model strings. Users can explore these new previews using the recently introduced aliases: gemini-flash-latest and gemini-flash-lite-latest.
As Google continues to refine its Gemini models, the company remains committed to engaging with users to shape future updates. They plan to notify users two weeks in advance through email before any changes to model versions occur. Both Gemini 2.5 Flash and 2.5 Flash-Lite are still available for those seeking more stability in their applications, as the company pushes forward in enhancing the capabilities of the Gemini suite.
