Skip to main content
Back

Google launches Gemini 2.5 Computer Use for UI control

xAI debuts Grok Imagine v0.9, OpenAI ships GPT-5 Pro API, ElevenLabs UI agents, Google voice search, Cursor Plan Mode

October 8, 2025 4 min read

Hey

Welcome to AlphaSignal, the most read source of news by AI engineers and researchers.


Every day, we identify and summarize the top 1% of news, papers, models, and repos, so you’re always up to date.


Here’s today’s roundup:

Summary

Read time: 4 min 21 sec

Top News

▸ Google releases new AI model that controls web and mobile UIs

ACI.dev

▸ Govern AI actions securely with Gate22’s MCP function-level permissions

Signals

▸ xAI launches Grok Imagine v0.9 with improved video quality, motion

▸ ElevenLabs introduces open-source UI components for AI voice agents

▸ OpenAI makes GPT-5 Pro available in API for harder reasoning tasks

▸ Cursor adds Plan Mode to research codebases and create detailed plans

▸ Google Research unveils Speech-to-Retrieval for direct voice-based search

Clockwise

▸ Optimize calendars with Clockwise MCP’s AI-driven scheduling intelligence

Trending Repos

▸ LLMs-from-scratch: Build GPT-style transformer models completely from scratch

▸ cua: Run AI agents controlling full OS via containers

▸ Handy: Offline speech-to-text app with local transcription

Top Lectures

▸ Andrew Ng teaches new Python-based Agentic AI course

Top News

Google introduces Gemini 2.5 Computer Use, a Pro variant optimized for UI agent control

alpha_signal_image_2

Google released the Gemini 2.5 Computer Use model, a specialized variant of Gemini 2.5 Pro that powers agents capable of interacting directly with web and mobile user interfaces. It leads on web control benchmarks with lower latency than competing systems.


Key Features

  • UI control: Click, type, scroll, drag, and fill forms directly.

  • Loop execution: Operates in a screenshot-action loop for reliability.

  • Confirmation requests: Prompts users for approval before high-risk actions.

  • Optimized domains: Tuned for browsers, with strong early performance on mobile apps.

  • Safety guardrails: Built-in protections against misuse, scams, and prompt injections.

How It Works

  • Loop design: Screenshot + history → model predicts action → environment executes → loop continues.

  • Action set: Click, type, scroll, drag, dropdowns, forms, logins.

  • Confirmations: Requests user approval before high-risk actions.

  • Domain: Optimized for browsers, generalizes well to mobile UIs.

Benchmark Performance

Gemini 2.5 Computer Use was tested on specialized benchmarks for UI agents:

  • Live web tasks: Gemini 2.5 Computer Use leads with ~72% accuracy.

  • Complex multi-step navigation: scores highest among tested models.

  • Mobile app control: strong performance, surpasses prior baselines.

  • Latency: Completes tasks in ~225s, lowest at this accuracy tier.

  • Tester results: 50% faster workflows, +18% accuracy, 25% fewer test failures.

Early Use Cases

  • Google internal: UI testing, Firebase Agent, Project Mariner, AI Mode.

  • Poke.com: Faster assistants across messaging platforms.

  • Autotab: Improved reliability on parsing-heavy automation tasks.

  • Google Payments: Stabilized pipelines, 60% automatic recovery of failed tests.

Access & Availability

  • Public preview: Available in Gemini API via AI Studio and Vertex AI.

  • Try demos: Browserbase offers immediate test environment.

  • Build agents: Use Playwright locally or Browserbase on cloud VMs.

Try now
alpha_signal_image_3

Govern Every AI Action With Open Source Control in Minutes

Gate22 is an MCP gateway from ACI.dev that gives you full oversight of every tool and action your AI uses.

  • 100% open source, Apache-2.0 license, free to start on cloud

  • MCP function-level permissions for all agent actions

  • Condense any number of MCPs and tools to one auditable MCP endpoint

  • Audit and track every tool change with detailed logs

  • Easy integration with any agentic IDE or AI tool you use

Try Gate22 and star on GitHub now.

Try gate22

partner with us

Signals

xAI rolls out Imagine v0.9 with smoother visuals, expressive vocals, and advanced audio-video harmony

23,871 Likes

ElevenLabs launches open-source UI library with 22 customizable components for AI audio and voice agents

2,183 Likes

OpenAI ships GPT-5 Pro API for hardest problems, with deeper reasoning capabilities

1,932 Likes

Cursor presents Plan Mode, letting agents research codebases, draft plans, and run longer workflows

2,143 Likes

Google Research publishes Speech-to-Retrieval, enabling voice search without converting speech into text first

689 Likes

Clockwise MCP Solves Multi-Party Scheduling and Focus Time Management

Clockwise MCP is the first Model Context Protocol server for intelligent scheduling. Manage meetings, block focus time, and coordinate team availability with working hours, time zones, and personal preferences built in.


Start optimizing your calendar with AI-driven scheduling today.

Explore Clockwise MCP Today ↗️

Repos

LLMs-from-scratch

74,892 Stars

This repo guides you through coding a GPT-like large language model from scratch in PyTorch. You build attention, tokenizers, pretraining, and finetuning stages step by step.

cua

10,296 Stars

cua enables AI agents to control full operating systems in lightweight containers. It provides Computer and Agent SDKs, multi-OS support, and benchmarks for training, evaluation, and deploying desktop automation.

Handy

1,482 Likes

It is an offline, cross-platform speech-to-text tool built with Tauri (Rust + React). It transcribes voice locally using Whisper or Parakeet models and pastes text directly into active text fields.

Top Lecture

Andrew Ng teaches new Python-based Agentic AI course

5,836 Likes

alpha_signal_image_4

The Agentic AI course taught by Andrew Ng offers a structured introduction to building AI agents that operate beyond single prompts. You will learn how to design workflows where agents reflect on their outputs, plan multi-step tasks, and coordinate with specialized sub-agents.

The course uses raw Python, keeping all implementation details visible without relying on external frameworks.


You will learn:

  • Reflection: how an agent reviews and improves its own outputs

  • Tool use: integrating APIs, databases, search, and code execution

  • Planning: breaking complex tasks into structured sub-tasks

  • Multi-agent collaboration: coordinating specialized agents for efficiency

  • Evaluation: using error analysis and metrics to refine workflows

  • Capstone: constructing a research agent that searches, synthesizes, and reports

Start now

At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We’re passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.


Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We’re based in the United States. By using our site, you agree to be governed by U.S. laws.

Looking to promote your company, product, service, or event to 250,000+ AI developers?

Work with us