Released in December 2025, the Gemini 2.5 Pro update introduced significant enhancements to Google's flagship multimodal model, specifically targeting its speech and audio processing capabilities. Built on a sparse mixture-of-experts (MoE) architecture, the model is designed for complex reasoning across text, code, images, and audio, featuring a context window of up to 1 million tokens.
The December 2025 update focused on the model's Text-to-Speech (TTS) performance, introducing improved expressivity, more precise pacing, and better multi-speaker dialogue consistency. These improvements allow the model to better adhere to style prompts and adjust tone and speed based on the provided context, facilitating more natural sounding interactions for applications like audiobooks and podcasts.
In addition to speech synthesis, Gemini 2.5 Pro maintains native audio understanding, enabling the transcription, summarization, and translation of audio files up to approximately 8.4 hours in length. The model operates within a unified framework that processes audio tokens alongside other modalities, allowing for cross-modal reasoning without the need for external transcription tools.