---

On June 5, 2025, Google DeepMind launched Gemini 2.5, its latest AI model designed to reason across text, images, and structured data in real time. The new model powers live transcription, contextual image understanding, and spreadsheet logic in Google Workspace, with latency reduced by over 50% compared to Gemini 1.5.

“Gemini 2.5 represents a leap toward native multimodality—understanding not just what you type, but what you see and organize,” said Demis Hassabis, CEO of DeepMind.¹

Gemini 2.5 includes a vision-language reasoning module that supports pixel-level comprehension and chart analysis. It also handles structured data, enabling queries like “what’s the revenue delta between Q2 and Q3?” with semantic accuracy.

Why it matters now

• Enterprises need AI that bridges unstructured input (images) and structured output (tables).
• Gemini 2.5 reduces friction across modality boundaries.
• Google sets a new bar for productivity suite AI with Workspace-native deployment.

Call-out: A model that sees your spreadsheet and understands your slide

Gemini 2.5 cuts latency by 51% and boosts visual question-answering accuracy by 37% over its predecessor.

Business implications

• Knowledge workers can query visual and tabular data in natural language.
• Finance teams gain new tools for visual audits and consistency checks.
• Developers can now access Gemini Vision APIs through Vertex AI.

Gemini 2.5 is rolling out now via Google Workspace Enterprise Plus, integrated into Duet AI for Sheets and Slides. Gemini APIs are available in preview for custom apps.

Looking ahead

DeepMind’s roadmap includes spatial reasoning and video chain-of-thought in Gemini 3.0. Google is also testing Gemini on Android and AR environments.

Forrester forecasts that by 2027, 65% of enterprise productivity platforms will feature multimodal AI tools—up from 18% in 2024.

The upshot: Gemini 2.5 brings real-time, multimodal intelligence to enterprise workflows, transforming AI from a writing tool into a perceptive digital assistant.

––––––––––––––––––––––––––––
¹ Demis Hassabis, Google DeepMind Announcement, June 5, 2025.