Google has announced updates for its Gemini 2.5 Flash and Flash-Lite models, releasing improvements for both the public-facing Gemini app and for developers through new preview versions.
According to a report by 9to5Google, users of the Gemini app will notice several enhancements to the Gemini 2.5 Flash model. Responses are now better organized with improved formatting, using elements like headers and lists to make information easier to read. The model’s reasoning capabilities have also been improved, providing clearer, step-by-step explanations for complex subjects, such as homework questions. Furthermore, image understanding has been enhanced, allowing the app to process more detailed images and diagrams. Users can, for example, upload photos of notes and ask Gemini to summarize them or create flashcards.
In a parallel announcement, Google released new preview versions of Gemini 2.5 Flash and Flash-Lite for developers, which are available in AI Studio and Vertex AI. The company stated that these releases are intended for testing and feedback to help shape future stable versions.
Improvements in Developer Previews
Google outlined specific improvements for each new developer model.
- Gemini 2.5 Flash-Lite (`gemini-2.5-flash-lite-preview-09-2025`)
- Instruction Following: The model is better at following complex instructions and system prompts.
- Conciseness: It produces less verbose answers, which can reduce token costs and latency.
- Multimodal Skills: The update includes more accurate audio transcription, better image understanding, and improved translation quality.
- Gemini 2.5 Flash (`gemini-2.5-flash-preview-09-2025`)
- Tool Use: The model has improved performance in complex, multi-step applications that use tools. Google claims a 5% gain on the SWE-Bench Verified benchmark compared to the last release.
- Efficiency: It is now more cost-efficient, achieving higher quality with fewer tokens.