Google releases Gemini 2.5 Flash and Lite previews with improved instructions & multimodal

Google has updated the Google Gemini app with improvements to the 2.5 Flash model. Responses now use clearer formatting with headers, lists, and tables, while homework help provides stronger step-by-step reasoning. The model also interprets images and diagrams more effectively, letting users upload notes for summarization, organization, or flashcard creation.

Developers can access preview versions of Gemini 2.5 Flash and 2.5 Flash-Lite through Google AI Studio and Vertex AI. These previews are not stable releases but aim to shape future versions. The 2.5 Flash preview doubles token efficiency, reducing costs and latency while improving quality. It also shows better agentic tool use, with a five-point gain on the SWE-Bench Verified benchmark. With “thinking” enabled, it produces higher-quality outputs using fewer tokens.

The Flash-Lite preview improves instruction following and generates more concise responses, lowering verbosity and costs. It also strengthens multimodal and translation features, including better image understanding, audio transcription, and translation accuracy.