Google launches updated Gemini 2.5 Flash and Flash-Lite with improved quality and token efficiency
Google has launched updated versions of Gemini 2.5 Flash and 2.5 Flash-Lite with improve performance, quality and token efficiency. These two models are now available on Google AI Studio and Vertex AI.
In terms of benchmarks, the Gemini 2.5 Flash and Flash-Lite has acheived major feat in the Artificial Analysis Intelligence index against their predated models for end-to-end response time and intelligence. On the other hand, Google has nearly doubled the output tokens efficiency for Gemini 2.5 Flash-Lite and around 24% reduction for Gemini 2.5 Flash.
Updated Gemini 2.5 Flash models Benchmark
Updated Gemini 2.5 Flash-Lite
The new model brings better instruction following for complex instructions and system prompts. It focuses on concise answers with lower token intakes and latency for high-throughout applications. This update enables improves audio transcription, image analysis and translation quality.
Updated Gemini 2.5 Flash
This model is getting an upgrade for its agentic skills, It can better select tools, enhance performance for complex tasks and more. On SWE-Bench, the model fetched 5% over its predecessor.
Access Improvements
Google has also improved latest model access, which eases worries for rememering speific model string. Instead, you can now put -latest alias for each model family to fetch its access.
This alias will also direct you to the most recent model versions without requiring you to update the string and keep your project updated with new model’s features. Here is the string example.
- gemini-flash-latest
- gemini-flash-lite-latest
Additionally, Google will send 2-week notice via email before making any updates or remove any model version behind – latest. However, you can always be specific if your project requires model stability.
Here are the stringsg for the updated Gemini 2.5 Flash and Flash-Lite:
- gemini-2.5-flash-lite-preview-09-2025
- gemini-2.5-flash-preview-09-2025
(source)