Published • loading... • Updated
‘Gemini 2.5 Computer Use’ Model Enters Preview with Strong Web, Android Performance
Google's Gemini 2.5 model enables AI agents to perform 13 browser actions with high accuracy and low latency, improving automation in UI testing and task completion.
- Starting today, Google LLC announced Gemini 2.5 Computer Use, a specialized model powering agents to interact with UIs via browser, available in public preview through Gemini API on Google AI Studio and Vertex AI.
- To address interface-only workflows, Google built the model so agents can perform tasks inside graphical user interfaces designed for human users rather than APIs.
- It processes three main inputs—the user request, a screenshot, and recent actions—and generates function-call outputs like clicking or typing in an iterative loop with screenshot and URL feedback.
- Google has deployed the model to production for UI testing, reporting automated repairs for over 60% of failing tests and adding safety features plus developer controls for risky steps.
- The announcement followed OpenAI's Dev Day by one day, amid prior competition from Anthropic PBC which released a `computer use` model last year, while Google reports strong Browserbase benchmark results of 65.7% and 79.9%.
Insights by Ground AI
22 Articles
22 Articles
Google's Gemini 2.5 Computer Use model powers AI agents to interact with UIs (GOOG:NASDAQ)
Google (GOOG) (GOOGL) has released the Gemini 2.5 Computer Use model, which is now available for developers to access via the Gemini API, the company announced on its blog on Tuesday.
·United States
Read Full ArticleFor now, AI models reach as far as the APIs allow: to overcome this limit, Google's Computer Use arrives, which scrolls pages, clicks and types like a user - on macitynet.it Gemini 2.5 Computer Use navigates and uses the browser like you do
Coverage Details
Total News Sources22
Leaning Left3Leaning Right1Center1Last UpdatedBias Distribution60% Left
Bias Distribution
- 60% of the sources lean Left
60% Left
L 60%
C 20%
R 20%
Factuality
To view factuality data please Upgrade to Premium