Google has widened its Gemini artificial-intelligence portfolio by releasing stable versions of the 2.5 Flash and 2.5 Pro models and by introducing Gemini 2.5 Flash-Lite in preview. Announced on 17 June, Flash-Lite is billed as the company’s fastest and most cost-efficient model, aimed at high-volume, latency-sensitive jobs such as translation and data classification. The model retains a one-million-token context window and allows developers to toggle more intensive reasoning tools depending on budget and task requirements. Google says Flash-Lite delivers up to 1.5 times the speed of earlier iterations while improving performance in maths, coding and multimodal benchmarks. In the AIME 2025 evaluation, it scored 63.1 percent, edging the standard Flash model and trailing the higher-end Pro’s 88 percent. Flash-Lite is immediately accessible through Google AI Studio and Vertex AI, while the now-generally-available Flash and Pro versions can also be used in the consumer Gemini app. Separately, the company has begun emailing Android users to confirm that, from 7 July, the Gemini assistant will gain broader on-device permissions. The update will let the AI interact directly with WhatsApp, Messages, phone utilities and Google services such as Maps, Calendar, Keep and Tasks, all within the voice-driven Gemini Live interface. The rollout includes real-time captions and new privacy controls that restrict notification and message access unless explicitly enabled by the user. The dual moves deepen Google’s push to challenge OpenAI and Anthropic by pairing a tiered family of language models with tighter integration across its mobile ecosystem, aiming to offer faster responses at lower cost while embedding AI more deeply into everyday Android workflows.