Quick Overview
This video is an event recap and product review presented by Claire on the How I AI channel following OpenAI DevDay in San Francisco. It examines the major keynote announcements, platform additions, and new model releases while testing early software builds. The presentation aims to help software engineers and technology teams identify which OpenAI updates are relevant to their workflows.
Key Points
- 1.OpenAI introduced Dots, which are persistent, always-on AI agents capable of operating across ChatGPT, Codex, local or cloud virtual machines, Slack, and voice calls.
- 2.ChatGPT Space launched as an AI-native collaborative workspace that allows teams and agents to co-edit pages, slides, and shared documents with integrated enterprise permission controls.
- 3.Sites now supports bundling data connectors and plugins directly into hosted web apps, enforcing user-level data access permissions automatically.
- 4.OpenAI announced the GPT-6.1 Sol model, offering near-Astra intelligence at one-fifth the cost, priced at two dollars per million input tokens and ten dollars per million output tokens.
- 5.The new Decisions API provides fast, constrained decision-making powered by the Luna model with integrated computer vision capabilities.
- 6.Astra Ultrafast provides inference speeds up to eight times faster at up to 350 tokens per second, enabling real-time interactive vector drawing and dynamic 3D world manipulation.
Summary
The presentation opens with a recap of OpenAI DevDay in San Francisco, highlighting major announcements across agent frameworks, collaborative platforms, and new frontier models. The central consumer and workplace announcement is Dots, an always-on, long-running AI agent integrated into ChatGPT, Codex, and Slack. Users on Pro, Business, and Enterprise plans receive an assigned Dot that can be reached via chat, phone-like audio calls, or Slack mentions. Each Dot runs within its own dedicated cloud virtual machine loaded with standard applications and browsers, but it can also be granted access to a user's local machine to execute background workflows, track tasks, and suggest proactive actions.
OpenAI also unveiled ChatGPT Space, an AI-native collaboration platform designed for teams and agents to jointly create and edit artifacts, documents, and slides. Space acts as a shared canvas that tracks notes, to-do lists, and project plans, allowing users to direct agents via inline mentions to update specific content blocks. In parallel, OpenAI expanded its Sites web application builder by adding support for bundled plugins and data connectors. Sites integrated with connectors such as Snowflake dynamically respect end-user credentials, restricting data visibility if a collaborator lacks administrative access.
Turning to developer infrastructure and model upgrades, the video highlights GPT-6.1 Sol, positioned as a workhorse model delivering intelligence close to the flagship Astra model at one-fifth the price. Sol is priced at two dollars per million input tokens, ten dollars per million output tokens, and ten cents per million cached input tokens, compared to Astra at ten dollars input and fifty dollars output. OpenAI also introduced the Decisions API, a low-latency endpoint pairing the lightweight Luna model with computer vision. A demonstration shows the API processing one hundred video frames in seconds to select optimal thumbnail images without awkward facial expressions, as well as powering binary visual classification tasks.
The most technically demanding release showcased is Astra Ultrafast, an inference speed tier operating eight times faster than baseline models, reaching speeds up to 350 tokens per second at six times the normal token price. Real-time applications built on this tier include The Other Pencil, an AI-assisted sketchpad where the model completes drawings into SVG vector graphics as the user draws, and Little Starship, an interactive 3D game environment where prompts dynamically generate 3D assets, alter scenery, and manipulate real-time physics rules.
Dots: Persistent Autonomous Agents
OpenAI announced Dots, long-running agentic assistants assigned to Pro, Business, and Enterprise tiers. Each user receives a dedicated Dot that operates across desktop, mobile, and Slack environments, with access to its own cloud virtual machine or the user's local machine to perform proactive tasks, schedule activities, and execute purchases.
ChatGPT Space and Enhanced Sites
ChatGPT Space introduces an environment where human teammates and AI agents collaborate on interactive documents, artifacts, and presentation slides. Alongside Space, OpenAI updated Sites to allow developers to embed live plugins and data connectors, ensuring that viewers only access underlying data sources for which they hold credentials.
New Models and Decisions API
OpenAI expanded its model catalog with GPT-6.1 Sol, delivering high reasoning capabilities at a fraction of the cost of flagship Astra models. Additionally, OpenAI previewed the Decisions API, a specialized tool combining constrained Luna intelligence with computer vision for sub-second classification and visual analysis tasks.
Astra Ultrafast and Real-Time AI Applications
The Astra Ultrafast tier delivers inference speeds up to 350 tokens per second, which is eight times faster than standard mode at six times the base cost. Demonstrations include The Other Pencil, a real-time collaborative SVG drawing canvas, and Little Starship, an interactive 3D environment capable of rendering live scene modifications and physics changes on the fly.
The Bottom Line
This recap outlines OpenAI's transition toward persistent autonomous agents, shared human-AI enterprise workspaces, and ultra-low-latency model inference. The presentation highlights the core technical trade-offs between speed, cost, and UI complexity across Dots, Space, and the Decisions API. While tools like Astra Ultrafast demonstrate new real-time capabilities, production viability for developers remains dependent on future pricing adjustments and information architecture refinements.
FAQ
What is OpenAI DevDay 2026 and what major product announcements were introduced at the event?
OpenAI DevDay 2026 is OpenAI's developer conference where the company announced persistent agentic Dots, the ChatGPT Space collaborative workspace, the GPT-6.1 Sol model, the Decisions API with vision, and the high-speed Astra Ultrafast inference tier.
What are OpenAI Dots and how do persistent agents interact with user computers?
Dots are always-on, long-running AI agents assigned to users on paid plans that operate across ChatGPT, Codex, Slack, and voice calls. Each Dot can execute tasks inside its own dedicated cloud virtual machine or run directly on a user's local computer with granted permissions.
What is ChatGPT Space and how does collaborative document editing work with agents?
ChatGPT Space is a collaborative enterprise workspace where human teammates and AI agents co-create documents, slides, and web artifacts. Users can mention agents directly in pages to update project plans, generate slides, and manage task lists.
How does GPT-6.1 Sol pricing and performance compare to the flagship Astra model?
GPT-6.1 Sol delivers near-Astra performance at one-fifth the cost, priced at two dollars per million input tokens, ten dollars per million output tokens, and ten cents per million cached tokens, compared to Astra at ten dollars per million input tokens and fifty dollars per million output tokens.
What is the Decisions API and how does computer vision enhance its classification speed?
The Decisions API pairs the fast Luna model with computer vision to deliver rapid, constrained answers on predefined sets of choices. It can evaluate video frames and visual inputs in sub-second intervals to perform tasks like video thumbnail selection and object detection.
What is Astra Ultrafast and what speed and cost trade-offs does it introduce?
Astra Ultrafast runs up to eight times faster than standard inference, outputting up to 350 tokens per second. However, it costs six times as much as standard token pricing, making it suited for specialized real-time interactive and gaming applications.
Worth watching for
Software engineers, AI product managers, and enterprise technology leaders evaluating OpenAI's latest agent infrastructure, model tiers, and collaborative workspace tools.
- openai
- devday
- chatgpt
- ai-agents
- software-development
- gpt-6