Zengyi Qin

Introducing Lux, the most powerful and fastest Computer Use model, built by OpenAGI Foundation @agiopen_org Lux outperforms Google Gemini CUA, OpenAI Operator and Anthropic Claude on benchmark with 300 real-world tasks. Try our developer-friendly SDK to build powerful, real-world applications. 🧵

Watch our Computer Use model visually organize a chaotic local directory directly at the OS level. Most agents are trapped inside the browser. Lux operates natively across Mac and Windows, driving the actual desktop GUI. In this sequence, Lux isn't running a background Python script. It uses pure visual intelligence to execute the workflow: Perceives the UI: Visually scans the native file manager to identify file types and icons. Synthesizes Context: Uses semantic reasoning to categorize unstructured files. Drives the OS: Executes rapid, native GUI selections to sort them into folders. By optimizing how the model processes UI elements, this OS-level execution is faster, cheaper, and significantly more accurate than current SOTA computer use models. #OpenAGILabs #ComputerUse #Automation #SoftwareEngineering #AI #OS

Watch our Computer Use model visually navigate a complex Shopify dashboard to add a new product natively in the browser. Zero API integrations or CSV uploads required. In this sequence, Lux executes a standard e-commerce operations task entirely through the GUI: Visual Parsing: Scans the dense Shopify sidebar to locate and navigate to the 'Products' module. Interface Interaction: Identifies and triggers the 'Add product' action within the dynamic DOM. Data Synthesis: Maps the unstructured variables (Title, Description) into the correct rich-text input fields. State Execution: Completes the workflow by saving the entry and visually confirming the success state. Bypassing rigid backend integrations and driving workflows directly at the presentation layer is how automation actually scales. #OpenAGILabs #ComputerUse #SoftwareEngineering #Automation #AI #Shopify

Wanna see how we got Lux to be 10x cheaper and 3x faster at the same accuracy as SOTA models? Come here me speak at the inaugural ClickConf in San Francisco luma.com/nbib4oev?utm_c… Sign up now for $20 in Lux credits

Watch our Computer Use model visually organize an inbox and execute a multi-step workflow right in the browser. It interacts with the Gmail interface natively, zero backend integrations required. In this sequence, Lux executes a standard operations task entirely through the UI: Visual Navigation: Navigates the sidebar to trigger the label creation menu and types "invoices". Native Tooling: Uses the built-in search bar to filter the inbox for financial documents. Interface Interaction: Clicks the bulk-select checkboxes and navigates the top menu to apply the new label. This is true zero-shot visual execution. You give a command, and the agent drives the browser. #OpenAGILabs #ComputerUse #SoftwareEngineering #Automation #AI

Watch Lux translate an unstructured chat message directly into a labeled GitHub issue. Our Computer Use model executes this cross-app GUI workflow with zero API integrations. In this flow, Lux is doing more than just clicking coordinates. It reads the raw bug report from the chat interface, synthesizes the core technical context (iPhone 15, landscape UI overlap), and drives the browser to the repository. It drafts the issue using structured Markdown, outlines the reproduction steps, and then physically interacts with the DOM to locate and assign the "bug" label in the sidebar before submitting. This is how you bridge messy human communication and structured engineering trackers natively. #OpenAGILabs #ComputerUse #SoftwareEngineering #GitHub #DevOps

@shenbokui Very impressive! Great work

Excited to introduce Uni-1, our new multimodal model that *unifies* understanding and generation. TLDR: a team of ~15 researchers is going pound-for-pound with nano banana and gpt image 🧵

@michaelsshang Very impressive. A companion with independent life will reshape how users feel about it

𝐈𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐢𝐧𝐠 𝐒𝐨𝐮𝐥𝐋𝐢𝐧𝐤, 𝐀 𝐍𝐞𝐰 𝐄𝐫𝐚 𝐨𝐟 𝐂𝐨𝐦𝐩𝐚𝐧𝐢𝐨𝐧𝐬𝐡𝐢𝐩. A beautiful life, unfolding next to you. Try free at getsoullink.com

Social media management shouldn't require endless clicking. What if you could trigger your entire workflow with a single command? In this demo, Lux handles social account management autonomously. The instruction is simple: "Boost my pinned X post for 2 days at $20/day." Watch Lux visually navigate the profile, identify the pinned post, and execute the promotion flow perfectly. No APIs. No manual setup. Just give the command, and let the Agent execute. #MarketingAI #SocialMediaManagement #OpenAGILabs #Automation #Tech

Claude Cowork proves the interest in AI coworkers. Orion (meetorion.app) takes the autonomous working experience to another level: • Works with and enhances your experience with your favorite apps (messages, docs, emails, sheets, files, calendars…) and AI tools (ChatGPT, Claude Code, Gemini, Grok…) • Supervises workflows when you are away, from vibe coding to media creation • Coordinates AI agents across all your devices (macOS, windows, VM), each operating one device independently Your personal AI superpower. New surprising ways of vibe working every day - I just tried using Orion to add subtitles to the video and it worked! Watch it handle multiple tasks autonomously from a single cmd+k.

Current agent evals are broken! Most teams just run the agent on tasks and report success rate. But that collapses a GUI agent's real complexity into a single binary outcome. Here's how we think about building better evals at OpenAGI 🧵

In the podcast, we discussed how we trained the State-of-the-Art Computer Use models, and how it will impact the paradigm of AI OS. Thanks @mlopscommunity for hosting the podcast!

@agiopen_org So cool

Never write a To-Do list anymore. We built a Lux agent that parses meeting transcript, identifies specific deliverables, and formats them into a Notion schema. No rigid APIs. Just natural language control. You can use Lux to create such agents. Full workflow 👇

@agiopen_org Huge step toward truly autonomous assistants.

From inbox to calendar invite: fully autonomous! Watch the OpenAGI model detect a scheduling link in an email, click through, and transition to the Calendly UI to book the meeting. No APIs. No human handoff. Just pure computer use handling cross-application workflows. 👇

This Thursday I'm hosting an AMA on our Discord to talk more about our foundation computer model, Lux, and what cool applications you can build with it. Join us! discord.gg/U93G3tWC?event…

The AI run operating system. @qinzytech shows me how AI will control our computers and bring a new way of computing to our homes and businesses. Such an honor to have him in my home yesterday. It lets AI agents run Macintosh, Windows, and Linux boxes like a human can. But I see it as a bigger deal than that. It, and a few other companies, are laying down the groundwork for robots, and humans with AI glasses, or brain/computer interfaces, to run our world in a whole new way. agiopen.org

The new AI operating system. @qinzytech just visited my home. While talking with him I got those goosebumps like I did when I first saw Siri, Tesla, Matic Robots, Insta360 (I was the first to see all those). Today at 4 p.m. he will join @IrenaCronin and me on X’s audio spaces to talk about the new operating system he built that has deep implications on both enterprise and home. Join us then.

Hi Danny - Appreciate your feedback. For those prompts, you can try the model 'lux-thinker-1' instead of 'lux-actor-1'. If you use the actor, you will need to write the prompt in a more detailed way. For example, "Click 'Settings' then click 'Profile', and Type 'John Doe' in 'Name', and then click 'Save'"

@agiopen_org I've been trying it for basic things and it has failed ~80% of the time on basic things. app.capacities.io/home/bb964f60-…

Introducing Lux, the most powerful and fastest Computer Use model. Lux outperforms Google Gemini CUA, OpenAI Operator and Anthropic Claude on benchmark with 300 real-world tasks. Try our developer-friendly SDK to build powerful, real-world applications. 🧵

@realkieranlewis @agiopen_org It's $0.1/1M input tokens, $0.9/1M output tokens. You can try with this developer document developer.agiopen.org

@qinzytech @agiopen_org why is pricing so hard to find? can you share? looks promising - will try this weekend

Introducing Lux, the most powerful and fastest Computer Use model, built by OpenAGI Foundation @agiopen_org Lux outperforms Google Gemini CUA, OpenAI Operator and Anthropic Claude on benchmark with 300 real-world tasks. Try our developer-friendly SDK to build powerful, real-world applications. 🧵

@brishtiteveja @agiopen_org Thanks Andy!

@qinzytech @agiopen_org Just tried this. It works really well