AI Briefing#computer-use
The latest AI and developer news about #computer-use, with the original source and a short summary.
The latest AI and developer news about #computer-use, with the original source and a short summary.
The FP8-quantized release targets reduced repetitive reasoning and improved tool adherence for long-horizon tasks.
The integration allows AI agents to navigate non-API systems, reducing manual technician intervention for browser-based support tasks.
The tool supports macOS, Windows, and Linux by reading OS accessibility APIs to return JSON structures with unique refs for agent interaction, avoiding pixel-based screen matching.
The model achieves an average WorkBuddy score of 83.18 and reduces hallucination rates from 12.95% to 7.2% while maintaining pricing of US$0.90 per million input tokens.
The 4B model uses 13.0% fewer model calls and 25.4% fewer output tokens than comparable systems after online RL.
The new model offers 50% lower standard API pricing than GPT-5.6 Terra and GPT-6.1 Sol, while scoring 15 percentage points higher on the Agents' Last Exam benchmark at one-third the cost.
The model supports configurable reasoning, structured outputs, and tool calling, with a fast mode option for latency-sensitive work.
The update introduces a sandboxed in-app browser, rebuilt computer use controls, and a 40% reduction in pre-message context size.
GPT-6 Astra scored 55.0% on Ironclad research tasks, a 32% improvement over GPT-5.6 Sol, while reducing estimated time per attempt by 48%.
OpenAI has released a comprehensive model guide for the GPT-6 family, detailing strategies for managing context costs, selecting reasoning levels, and handling long-running workflows across GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna.
Cua has released CUA-S1, a small specialized model for computer-use decision-making, along with related open-source tools.
OpenAI has released GPT-6 Astra, significantly improving performance and alignment safety, including achieving 99.9% on ARC-AGI-3.
Meta has opened the iOS waitlist for its agent superapp 'Muse' and is testing a model variant that supports computer control capabilities.
Claude has released a beta feature that allows it to launch and control apps in the background on macOS desktops while users are working.
Anthropic officially launched its computer use and skills APIs, enhancing efficiency for production agents.
The update includes a new browser use tool and expands availability to Microsoft Foundry, with Vertex AI support coming soon.
Tencent has released UI-Mate-27B, an open-source GUI agent that performs desktop tasks using natural language instructions and demo-based learning.
AI agents have begun handling repetitive office tasks in real computer environments.
In the era of AI agents, UI does not disappear; instead, an AX layer for agents is added as a new tier.
Connect computer and browser usage to agents to automate bug reproduction and feature verification.
It points out that AI agents bypassing APIs instead of directly manipulating the GUI is a bottleneck for performance.
pi-computer-use, an MIT open-source extension that lets AI agents directly operate desktop apps without APIs, has been released.
Coasty has launched a managed API for AI agents that perform computer control and automation tasks.
A hardware KVM based on MCP has been released, helping AI agents directly control physical devices.