OpenAI Dev Day Recap: Dots, GPT-6.1 & More

This article may contain affiliate links. We may earn a commission if you purchase through them, at no extra cost to you.

OpenAI just threw a lot at the wall — and most of it stuck. The company's latest developer conference unveiled a slate of tools that go well beyond incremental updates: persistent AI agents called Dots, a new collaborative workspace called Spaces, two new model tiers (GPT-6.1 Soul and Ultrafast), a Decisions API, cloud-based Codex, and an OpenAI marketplace. That is a lot to process.

This review cuts through the noise. We examine what was actually announced, what the real-world implications are for individuals and businesses, where the pricing makes sense, and where you should pump the brakes before getting too excited. If you've ever wondered whether AI productivity tools are worth paying for — or if you're curious how these announcements affect the broader case for building AI into your workflow — read on.

What Are OpenAI Dots? The Agent Layer Explained

Dots are the headline product from this Dev Day, and they are genuinely different from what most people think of when they hear "AI assistant." A Dot is a persistent, always-on AI agent assigned to you personally. It is not a chatbot you open and close. It runs in the background, monitors your connected apps, takes initiative, and handles tasks without being explicitly prompted every time.

The demonstration showed a Dot named "Doy" doing things like:

• Detecting a rescheduled calendar event and updating the calendar automatically
• Scanning overnight user-testing feedback and surfacing a design change
• Building a working iPhone app prototype using Codex on a local laptop
• Delegating tasks in Slack on behalf of the user
• Fixing bugs by picking up session IDs from a feedback channel and opening pull requests

That last point is significant. Engineers at OpenAI reportedly had their Dots fixing dozens of bugs per day autonomously. This is not a party trick — it represents a genuine shift in how software teams could operate.

Dots are powered by Astra, OpenAI's most capable and safety-aligned model. You can set permissions, define boundaries on computer and app usage, and customise how your Dot behaves. It works across more than 4,000 apps and integrates with plugins you've already set up in ChatGPT.

Who gets access? Dots are available immediately to ChatGPT Pro, Business, Premium, and Enterprise users. Crucially, conversations with your Dot do not count against your usage limits — a smart move that removes a significant friction point.

ChatGPT Spaces: Collaborative AI Workspaces Worth Your Attention

Spaces is OpenAI's answer to collaborative productivity — think Notion meets Google Docs, but with AI agents as active participants rather than passive tools.

Inside a Space, you can:

• Write plans, generate images, build interactive charts, and create live prototypes
• Tag teammates and Dots in the same page
• Give the Space standing instructions (e.g., "check the API platform Slack channel and update this page daily")
• Build and share presentations collaboratively — including with your teammates' Dots

The presentation feature is listed as "coming soon," so temper expectations there. But the core collaborative page functionality is live today for eligible users.

The practical upshot: Spaces reduces the number of tool switches in a typical workday. Instead of bouncing between Notion, Figma, Slack, and a spreadsheet, you could theoretically run a product launch from a single Space. That is the pitch, anyway. Real-world adoption will determine whether the integrations are tight enough to justify consolidation.

GPT-6.1 Soul and Ultrafast: The Model Pricing Breakdown

This is where things get genuinely interesting for anyone thinking about cost efficiency — including developers running agents at scale.

GPT-6.1 Soul

GPT-6.1 Soul is positioned as a workhorse model: near-Astra-level intelligence at one-fifth of the price. The headline number is a 95% discount on standard cached input costs. For agents that repeatedly reference long contexts, iterate on code, or run multi-step tasks, this is a meaningful reduction.

OpenAI's framing is direct: Soul is the model you run most of the time so you can save Astra for tasks that genuinely need it. That is a sensible tiered approach, and the economics work if your use case involves high-volume agent tasks.

Ultrafast

Ultrafast is a different product entirely. It delivers eight times the speed of the standard model at six times the price — or twice the speed of Fast at twice the price of Standard. The demo showed it generating and launching a working rocket-themed app in the time it took the standard model to render the same output barely halfway.

The benchmark cited: 300 tokens per second. OpenAI claims there is nothing comparable on the market currently. That claim is credible for now, though competitors like Anthropic and Google are not standing still.

Ultrafast is available now for Astra and will be extended to GPT-6.1 Soul. The Decisions API — which enables near-instant model responses by giving the model a predefined set of choices — complements Ultrafast for routing, classification, and agent decision-making tasks.

Pros and Cons: Honest Assessment

Pros

• Dots are genuinely novel — persistent, permissioned, context-aware agents are a step beyond chatbots
• GPT-6.1 Soul pricing is aggressive — 95% discount on cached input makes high-volume agentic workflows economically viable
• Ultrafast is fast — 300 tokens per second is class-leading and practically useful for real-time applications
• Dots don't consume usage quotas — removes a major adoption barrier for power users
• OpenAI Marketplace gives developers a new distribution channel with 30+ launch partners on day one
• Codex is now fully cloud-hosted — removes local setup friction for development teams
• Spaces is built for teams — not just individuals, which broadens the addressable market significantly

Cons

• Dots require significant data access to be useful — connecting email, Slack, calendar, and code repositories raises legitimate privacy questions that OpenAI addressed only briefly
• Specialist Dots for enterprise are preview-only — the most powerful organisational use cases are not self-serve yet
• Ultrafast costs six times the standard price — for high-volume users, that adds up fast
• Spaces presentation feature is "coming soon" — the collaborative layer is not fully baked
• The Decisions API is narrowly scoped — it excels at routing and classification but is not a general-purpose speed solution
• Heavy dependency on ChatGPT ecosystem — businesses not already embedded in OpenAI's platform face a meaningful switching cost to adopt any of this

Comparison: Where OpenAI Dots Sit in the AI Agent Market

OpenAI's edge is the breadth of integrations and the autonomy of execution — particularly the code-writing and bug-fixing capability. Google's strength remains Workspace depth. Microsoft Copilot is the pragmatic choice for organisations already in the M365 ecosystem (and notably, OpenAI is integrating Specialist Dots into Microsoft 365 Agents, which softens the either/or choice).

OpenAI Marketplace and the Business Case

The OpenAI Marketplace is a smart commercial move. With 30+ launch partners including Code Rabbit, Notion, and Vercel, OpenAI is building a distribution flywheel: enterprise customers with existing OpenAI spending commitments can apply that budget to third-party tools in the marketplace. This lowers the barrier for developers to reach customers and gives enterprise buyers a reason to consolidate spend with OpenAI.

For startups and indie developers: this is a new customer acquisition channel worth taking seriously. If your product solves a problem that ChatGPT Enterprise users have, the marketplace removes one layer of the sales cycle.

For enterprise buyers: the ability to use existing OpenAI commitments in the marketplace is a genuine procurement advantage. It simplifies vendor management and can accelerate budget approval for AI tooling.

Overall Verdict

Rating: 8.5 / 10

OpenAI Dev Day delivered substance, not just slides. Dots are the most credible attempt yet to make AI agents practical for knowledge workers and engineering teams. GPT-6.1 Soul's pricing makes agentic workflows economically defensible. Ultrafast's speed benchmark is real and meaningful for real-time applications.

The caveats are real too. Privacy and data access questions around Dots deserve harder answers than a passing mention. Several headline features — Specialist Dots, the Spaces presentation layer — are previews, not products. And the Ultrafast pricing tier will sting at scale.

But for teams already on ChatGPT Business or Enterprise plans? The case to start experimenting with Dots immediately is strong. The cost is zero (it is included in your plan). The upside — even at a fraction of the autonomy shown in the demos — is substantial.

Bottom line: OpenAI has moved the goalposts on what an AI productivity platform looks like. The question is no longer whether AI agents can do useful work. It is whether your team's workflows and data governance are ready to let them.

Frequently Asked Questions

What is an OpenAI Dot and how is it different from ChatGPT?

A Dot is a persistent AI agent that runs continuously in the background of your work. Unlike ChatGPT, which requires you to initiate every conversation, a Dot monitors your connected apps, takes initiative, flags issues, and completes tasks autonomously. It retains context about how you work over time and can act across Slack, email, calendars, browsers, and codebases — without you being present in a chat window.

How much does it cost to use Dots and GPT-6.1 Soul?

Dots are included at no extra cost in ChatGPT Pro, Business, Premium, and Enterprise plans. Conversations with your Dot do not count toward usage limits. GPT-6.1 Soul offers a 95% discount on standard cached input pricing, making it significantly cheaper than Astra for high-volume or repetitive agent tasks. Ultrafast is priced at six times the standard model rate — premium, but justified for latency-sensitive applications.

Who should NOT invest time in setting up OpenAI Dots yet?

If your organisation has strict data governance policies, limited tolerance for AI accessing internal systems, or has not yet established clear AI usage guidelines, Dots are not ready for you. The agent requires broad data access to be genuinely useful — email, Slack, code repositories, calendars — and that access needs to be granted deliberately and governed carefully. Organisations in regulated industries (financial services, healthcare, legal) should also wait for clearer compliance documentation before connecting sensitive systems.

How does OpenAI Ultrafast compare to other fast AI models on the market?

Ultrafast delivers approximately 300 tokens per second, which OpenAI claims is unmatched currently. For context, most frontier models operate in the 30–80 tokens per second range under typical load. The practical difference is visible in real-time applications: code generation, agentic decision loops, and interactive prototyping all benefit meaningfully from this speed. The Decisions API extends this further by reducing latency for routing and classification tasks to near-instant response times. Competitors including Anthropic and Google DeepMind are actively working on speed improvements, so this lead may not be permanent — but it is real today.