OpenAI DevDay 2026 Keynote (FULL)

Sam Altman kicked off OpenAI’s DevDay by highlighting the rapid progress driven by user feedback, announcing a suite of new products designed to empower developers and users alike. The overarching theme emphasized AI as a transformative force, akin to a Renaissance, focused on augmenting human capabilities rather than merely industrial efficiency.

New Ways to Work with AI: Personal Agents & Collaboration

Dots (Personal AI Agents)

OpenAI introduced Dots, described as “remarkably capable, always-on agents built to handle anything you can think of.” These agents are envisioned as personal helpers, working 24/7 to offload tasks, allowing users to focus on higher-level work. Key features and capabilities include:

• Delegation: Dots can handle a wide range of tasks, from mundane (booking tickets, dinner reservations) to ambitious software development. An example given was delegating the task: “Get the app off legacy API before it shuts down,” where a Dot can trace dependencies, write code, run tests, and create pull requests.
• Technical Foundation: Dots operate in a cloud environment, have their “own computer,” can write and test code, and integrate with existing user plugins. This allows them to follow work across various platforms and applications.
• Future Interaction: Beyond ChatGPT, Dots will soon be accessible via text messages and phone calls, providing updates and soliciting input.
• Safety and Control: Powered by Astra, Dots incorporate built-in protections and safeguards, with customizable instructions, allowing users to control the level of responsibility delegated.

ChatGPT Spaces (Collaboration Platform)

Recognizing that most existing productivity software isn’t built for effective AI-human collaboration, OpenAI launched ChatGPT Spaces. This platform is designed for teams to work together, with Dots as integral collaborators:

• Direct Integration: Dots can work alongside users directly within a page, track dependencies, and update findings. Users can tag a Dot in comments to assign tasks.
• Dynamic Collaboration: Holly Li demonstrated how Dots, like her personal agent “Dotty,” manage schedules, review feedback, and even generate code. Spaces support interactive charts, sheets, and live prototypes, where Dots can keep visualizations updated hourly.
• Team Amplification: Dots can locate information from private DMs, incorporate feedback, and even proactively fix bugs. Engineers at OpenAI are using their Dots to fix dozens of bugs daily, illustrating their role as delegating team members.
• Availability: Dots and Spaces are available to ChatGPT Pro, Business, Premium, and Enterprise users. Enterprise users also gain access to Specialist Dots, virtual teammates capable of scaling intensive work and improving with feedback, integrated with Microsoft tools.

Platform for Building: Models, Tools, Distribution

OpenAI is committed to providing developers with the best models, tools, and distribution channels.

Models for Developers

Responding to demand for cheaper and faster models, OpenAI announced significant updates:

• ChatGPT 6.1 Sol: A powerful new model offering a “great balance of cost and capability,” providing near Astra-level intelligence. It’s designed to be more cost-effective for context reuse, iteration, and longer tasks.
• Astra Ultrafast: A new speed tier offering 10 times the speed of standard models, reaching an “incredible 300 tokens per second.” This significantly boosts development and response times. The Pro-500 subscription will leverage UltraFast, offering 25x speed.
• Decisions API: This API allows models to respond in a fraction of a second by focusing on a predefined set of options, enabling extremely fast, almost instantaneous decision-making.
• One-Stop Shop: OpenAI aims to be a comprehensive provider across intelligence, price, speed, voice, images, and frontier models.

Accelerating Research with AI

Tejal Patwardhan highlighted how OpenAI’s models are transforming research:

• AI Research Intern: OpenAI achieved its bold prediction of having an AI research intern, showcasing the advanced capabilities of their models in areas like math, coding, and computer use.
• Improved Training & Safety: Models are used in a continuous loop to improve their own training, leading to safer interactions and reduced errors in tasks like desktop navigation. This has resulted in more than 2x the models shipped.
• Research Automation: Models can now take on over a third of day-long research tasks without human intervention, accelerating scientific discovery in fields such as solving math problems, antibiotic development, ancient language study, energy forecasting, and robotics.

Enhanced Developer Tools

OpenAI is making its internal tools accessible to developers:

• Codex Harness: The underlying technology powering Codex and Dots is now open-sourced and available in the cloud, allowing developers to start tasks on one device and continue on another.
• No-Code Cloud: A new cloud environment for hardening software, featuring deduplication, scheduled scans, and an improved interface.
• New API with Computer Use: This API provides access to the harness and multi-agent controls, allowing agents to interact with computer interfaces, such as navigating browsers and testing websites.
• OpenAI Private Attention (ZD.R.): A new privacy offering ensuring frontier model safety without storing user content on OpenAI’s servers, setting a new standard for data protection.
• API Reliability and Speed: OpenAI has achieved over 99% reliability and cut the time to first token by 45%, making their API highly performant.

Distribution and Monetization

To help developers build and grow businesses, OpenAI is leveraging its vast user base:

• Sign in with ChatGPT: With 1.2 billion ChatGPT users, developers can enable users to sign into their apps and utilize their existing ChatGPT plan tokens, simplifying access and reducing user acquisition friction.
• Plugin Extensions: Developers can now build plugins that integrate directly into ChatGPT and Codex, creating editors, dashboards, and workspaces. These plugins can be discovered within conversation flows when relevant.
• Simplified Submission: The plugin submission process has been streamlined to allow easier human review.
• Partnerships: OpenAI is fostering monetization through 30+ launch partners and enabling enterprise customers to leverage their existing plans, including a partnership with Base Ten for open-source model access.

Live Demos by Romain Huet

Romain Huet showcased the new capabilities through several impressive live demonstrations:

• Real-time 3D World Interaction: Using UltraFast, he demonstrated changing elements in a 3D environment, such as seating “dots” and displaying a live stream on a virtual screen, almost as fast as thought.
• Command-Line App Development: A React app was built and modified in real-time using UltraFast via the command line, highlighting GPT live’s ability to react instantly.
• Astra Playing a Game: Astra successfully played a 3D space game designed for humans, learning its dynamics and initiating game actions, demonstrating its ability to use any application.
• App Auditing with App Shots: An app screenshot was used to prompt Codex to audit the app across different screen sizes, with Codex navigating the computer, opening simulators, and testing features autonomously.
• Cloud-based Backend Rewrite: An ambitious task of rewriting an entire backend in Rust was delegated to the cloud, emphasizing asynchronous, hands-off development.
• Multimodal Robot Interaction: A robot, integrated with OpenAI’s API, demonstrated multimodal capabilities by drawing the audience, quacking on command, and generating a photo, showcasing hardware integration and real-time perception for robotic actions.

Sam Altman concluded by reiterating the vision of AI as a “Renaissance,” empowering individuals and fostering creativity, rather than merely driving industrial efficiency. The announcements aimed to provide the tools for developers and users to build a future where human potential is vastly expanded.

• Tags:
• AI Agents
• Developer Tools
• API
• AGI
• Collaboration
• DevDay
• Decisions API
• Dots
• ChatGPT Spaces
• GPT-6.1 Sol
• Astra Ultrafast
• Codex in Cloud
• OpenAI Private Attention
• Plugin Extensions