Tag
Vantage is an open-source Windows 11 app that turns a typed prompt into autonomous desktop actions, using AI models to read actual window text via UI Automation, click, type, and run PowerShell rather than relying on screenshots. It brings your own API key for any OpenAI-compatible provider, stores everything locally, and is an early-stage project still seeking failure-case feedback.
The Claude API has increased the per-request size limit for Messages and Token Counting from 32MB to 128MB, roughly a fourfold increase, to support tasks such as long conversations containing many images and large PDFs.
Vercel Labs has open-sourced agent-desktop, enabling any coding agent to gain desktop-level Computer Use capabilities by directly reading the system accessibility tree rather than screenshots. It supports background operation and simultaneous control by multiple agents, and is compatible with macOS, Windows, and Linux.
A tweet arguing that while Astra 6 is an older model, running inside the Codex harness gives it a level of AGI-like capability other frontier models lack, with seamless computer and internet use.
OpenAI released GPT-6.1 Terra, a midrange model emphasizing visual reasoning, computer use, and clearer writing, priced at $1/M input and $5/M output tokens — roughly half the cost of GPT-6.1 Sol.
Greptile is leveraging GPT-6 Astra's computer use capabilities to automate application testing beyond static code review, alongside its Daybreak security product and Codex-integrated Knowledge Base, in a push to reduce bugs and rework for developers.
GPT-6.1 Terra 发布，OpenAI 称其在快速、能干的计算机使用（computer use）方面实现新突破，能理解屏幕内容、跨应用执行操作，并以更低价格提供更高效的工作能力。
Cursor now lets AI agents control desktop applications like the iOS Simulator, enabling them to test the features they build directly on the machine.
OpenAI 发布 GPT-6.1 Terra，一款兼顾视觉推理、计算机操作与自然写作的高效模型，API 价格较 GPT-5.6 Terra 低 50%，在 Agents' Last Exam、MMMU-Pro、OSWorld 2.1 和 Terminal-Bench 4.0 等基准上大幅提升，同时显著降低幻觉与事实错误率。
作者开源了一个 Mac 应用 + MCP 服务器 Open iPhone Use，让 AI 智能体通过屏幕共享和蓝牙键鼠控制真实 iPhone。支持 ChatGPT/Codex、Claude Code 等任意支持 computer-use 的智能体，仅需 iPhone 和 Apple Silicon Mac。
Opposable 推出可在 Android 手机本地运行的 GUI 智能体：使用 4-bit 量化的 GUI-Owl 1.5 2B 模型（基于 Qwen3-VL，Alibaba Mobile-Agent 团队出品），通过 llama.cpp 在设备上直接读屏并操作任意应用，无需服务器。通过裁剪状态栏、丢弃平坦背景图像 token 等优化，Mac M2 CPU 上单步耗时从 10.6 秒降至 7.3 秒，且在 36 屏测试集上精度不降。
The author argues that beyond agent harnesses, agent interfaces are a crucial direction for self-evolving agents, enabling better efficiency, lower costs, and more reliable human-agent collaboration as seen with Artifacts. The post quotes DAIR.AI highlighting a paper on evolving agent interfaces as a trend to watch for agent builders.
Nace releases Drex Max 2.0, a 27B multimodal decision model for computer-use agents that scores options in a single forward pass with millisecond latency, priced at $0.06 per million input tokens via the Nace API.
PAPAV 提出了一个以能力为核心的统一框架（Perceive、Anticipate、Plan、Act、Verify），用于对比多模态具身智能体、计算机使用智能体与机器人系统，并基于 64 个基准揭示了 Anticipate 与 Verify 能力评估的显著缺口。
MiniMax 发布 M3.1 Flash 新模型，在复杂 agentic 任务、计算机使用、软件开发与网络安全等多类基准上全面提升，其中 DeepSWE v1.1 达到 73.7%，跻身顶尖模型行列。
MiniMax 发布 MiniMax M3.1 Flash，作为新一代 Flash 模型，在 agentic 任务、编码、Computer Use 与网络安全方面实现全面跃升，在线推理速度提升近三倍，使用成本降低 70%，输入/输出定价为每百万 token $0.15 / $0.60。
Nace AI 发布多模态决策模型 Drex Max，可直接从截图、屏幕录像、扫描文档和图表中快速做出决策，适用于 computer-use 和 browser-use 智能体。该模型在 Mind2Web 上达到 88.3%，Decision Index 得分 67.6，推理速度为 34ms/决策。
Someone has extracted Codex's Computer Use runtime from the ChatGPT desktop app into an open-source tool called LCU, letting agents like Claude Code, Pi, and Hermes see the screen, click buttons, and operate desktop apps — no Codex login required.
The paper presents MFM, a multimodal filesystem memory architecture that combines episodic, spatial, semantic, procedural, and working memory to help long-horizon embodied agents reuse experience, achieving improved performance on Minecraft and computer-use benchmarks across three base models.
Browser and computer use become significantly easier with updates to the latest OpenAI Python and Node.js SDKs, as highlighted by Kevin Whinnery in a threaded post.