Google DeepMind has announced Gemini 2.5 Ultra, a research-oriented variant of its flagship multimodal model. The new release is designed to compete with the latest reasoning models from OpenAI and Anthropic, with particular emphasis on scientific literature, mathematics, and long-document analysis.
Native Tool Use and Workspace Integration
Gemini 2.5 Ultra can call Google Search, Google Scholar, and internal enterprise knowledge bases as part of its reasoning chain. Users inside Google Workspace can ask the model to summarize a long Gmail thread, pull numbers from Sheets, and generate a Slides outline without leaving the sidebar. Google says this is the first Gemini model where tool use is native rather than orchestrated by a separate system.
The integration extends to Gmail filters, Calendar scheduling, and Drive permissions. Researchers can build reusable workflows that query documents, extract data tables, and update lab notebooks automatically. Google promises that enterprise data is never used to train the base model without explicit consent.
Benchmark Performance
On internal benchmarks, Gemini 2.5 Ultra reportedly scores above its predecessor on graduate-level science questions, coding interviews, and multilingual understanding. Google has also improved factuality grounding, showing inline citations for web-based answers and highlighting sources in Docs.
Availability
The model is available to Google One AI Premium subscribers, Cloud enterprise customers, and researchers through the Gemini API. Pricing is competitive with other frontier APIs, and Google is offering $300 in credits for new Cloud accounts that want to experiment with the model.
Industry Impact
Industry watchers view this announcement as another sign that the artificial intelligence market is shifting from raw capability demonstrations toward production-ready features. Buyers are increasingly focused on total cost of ownership, data governance, vendor transparency, and long-term support. The move also pressures competitors to respond quickly, which should accelerate innovation and drive more flexible pricing across the market. For end users, the practical result is likely to be better tools, clearer licensing terms, and stronger safety guardrails as the industry matures through 2025 and 2026. Enterprises that move early may capture meaningful workflow efficiencies before these capabilities become table stakes.