OpenAI Dev Day 2026: Everything Announced in 15 Minutes

Tech concepts, product features, and announced updates (high level)

1) Code X / Codex platform expansion

• Requested features were added:
  • Linux support for Code X
  • Mobile support (“for your phones”)
• Codex / Codex CLI updated with improvements (used later in a React/CLI demo)
• Codex moved to the cloud (“Codex is now completely in the cloud”), enabling faster, remote execution for building tasks and apps

2) “Dots” — always-on AI agents with memory of how you work

• Introduced as “extremely capable and always-on agents” designed to perform imagined tasks
• Personalization over time: Dots learn “how I work and what I want to focus on.”
• Delegation beyond small tasks:
  • Can handle simple workflows (booking, scheduling, errands-like tasks)
  • Can also handle “serious tasks” because it’s built on Astra and OpenAI’s underlying developments
• Astra-powered + security:
  • Powered by Astra (described as their most consistent model)
  • Includes built-in protection/security features
• Access control & limits:
  • Respects your access rights
  • You can set limits on app/computer usage and define instructions for actions
• Own cloud workspace:
  • Each Dot has its own “computer in the cloud,” including its own browser
  • Can write and test code as needed

3) How you communicate with Dots (multimodal + asynchronous messaging)

• ChatGPT is not required to interact with your Dot
• Planned/soon capability: text-message interaction
  • You send messages, and the Dot replies via text
• Dots can also:
  • Send updates/notifications and ask for opinions
  • Be called for direct voice interaction

4) Dot integration with existing tools/apps

• Works with 4,000+ apps in ChatGPT
• When you connect a Dot, it can access the same plugins you already use
• Dots can “follow your work” wherever it happens

5) “ChatGPT Space” — team workspace for plans, research, and agent collaboration

• Introduced as a shared environment for you, colleagues, and agents
• Core capabilities in the Space page:
  • Writing plans and conducting research
  • Creating images
  • Generating data
  • Building/prototyping interactive artifacts
• Agent collaboration inside Space:
  • Dots can be tagged in comments to start work immediately
  • Example instruction: check a Slack API channel and update results daily
  • A new way to create/share presentations within Space

6) “Dots” in real workflows (Slack, group chats, PRs, bug fixing)

• Positioned as practical collaborators inside day-to-day communications:
  • Forward Slack discussions to a Dot to complete tasks
  • Teams can treat Dots as representatives in group chats
• Demo-style examples:
  • Dot picks up session IDs/user errors from a feedback channel
  • Dot investigates and opens a PR with a fix
  • Claim: engineers fix “dozens of bugs every day” using Dottie

7) Hands-on development flow: build apps → run simulators → open PRs

• Live demo indicates:
  • Dots use Codex on the user’s laptop
  • They can build an app and run it in an iPhone simulator
  • Workflow shown: start from a phone mockup → local execution → generate a working app → continue sending PRs

8) Availability + specialized corporate Dots + Microsoft integration

• Availability: Dots and Space for ChatGPT Pro/Business/Premium/Enterprise, starting “today”
• Usage detail: calls don’t count against usage limits; Dots are included in plans and available 24/7
• Specialized Dots introduced:
  • Company-configured “virtual colleagues” for scaling labor-intensive work (accounting, marketing, law, etc.)
  • Companies set goals, provide context, and review results via feedback loops
• Microsoft partnership:
  • Integration with Agent 365, letting companies manage Dots through Microsoft tools

9) New model: “GPT 6.1 Soul” (cost/performance for agents)

• Claimed as near Astra-level intelligence at ~1/5 the price
• Strong agent usefulness: data entry costs 95% off vs standard
• Positioned as:
  • Cheaper context reuse
  • Better iteration
  • Longer task completion
• Mentioned as “smarter than Astra” (subtitles suggest incremental improvement) and positioned as a daily work tool

10) Ultrafast model via API + ChatGPT + Codex

• Performance claims:
  • “Fast”: 2x speed for 2x cost (relative)
  • “Ultrafast”: 8x speed for 6x cost vs standard
  • Speed cited: ~300 tokens/second
• Demo: side-by-side rocket creation shows Ultrafast finishing much faster
• Availability notes:
  • Ultrafast available with Astra now
  • Soon also with 6.1 Soul

11) “Solution APIs” + agent orchestration for speed and safety

• A “Solutions API” approach using predefined options/routing, such as:
  • Route request
  • Classify image
  • Decide next agent action
• Focus on decision/routing for extremely fast responses while maintaining:
  • image understanding
  • broad language support
  • security features
• Substrate described:
  • Starts with Codex framework, powering Codex Work and Dots
  • Optimized for accuracy, speed, and minimum tokens
• “Luna model” referenced (in subtitles) as part of the predefined-option routing approach

12) 3D/multimodal world updates in real time using agents

• Demo: interact with a room using a superfast model
• Example actions:
  • Add “dots” into the world
  • Turn camera to the hall and place a live broadcast on a big screen
  • Real-time changes described as happening as work progresses

13) Codex CLI + two-way audio and real-time terminal interaction

• CLI demo:
  • Create a React app that randomly selects participants, then expands selection size
• Two-way audio communication:
  • Terminal supports real-time audio response (not “ordinary dictation”)
  • Dot/bot reacts in real time; app generated quickly
• Demo ends with an in-browser connection/control scenario where the agent learns game behavior from the video/game interface

14) Hugging Face programmable robot demo (micro-duck)

• Hugging Face provides an early version of a programmable robot (“micro-duck” named Lavender)
• Combined modalities demo:
  • Audience interaction and robot responding (croak/acknowledgement)
  • Robot reacts to environment cues (e.g., “photo of all of you taken”)

15) OpenAI Marketplace launch

• Launch of an OpenAI marketplace for selling product integrations to companies
• Mentions 30+ partners (e.g., Code Rabbit, Notion, Vercel)
• Corporate customers can use part of their OpenAI commitments to buy products on the marketplace
• Open source models access via partnership with Base10

Main speakers / sources (as indicated in subtitles)

• Main presenter: “Today I’m going to show you…” (speaker at OpenAI DevDay; unnamed in subtitles)
• Fictional demo voice/user: “Dotty” / Dots agent (fictional music app demo “Blossom”)
• Characters mentioned: Holly (demo), Gabe, Patrick, Rennie, Rishabh, Laura
• Partner mentioned: Hugging Face team/representatives (robot demo); Microsoft (Agent 365 integration)
• No explicit real-person OpenAI CEO/CTO name captured in the provided subtitles