“Dots just gets it. Is it actually smarter than Codex?”
You might feel that way when one assistant carries a project forward and the other keeps asking questions, checking for problems and stopping before anything is delivered. You wanted to delegate a task. Somehow you became your AI's unpaid project manager.
You ordered help and received another management job. What a bargain.
Still, feeling smarter in daily use is different from proving superior reasoning ability.
Is Dots only available on Pro?
For individual subscriptions, essentially yes as of October 11, 2026. OpenAI's official guide describes a gradual rollout to eligible ChatGPT Pro subscribers, not Free, Go or Plus. Pro starts at US$100 per month.
It's not literally Pro-only: Business Premium also includes dots, and Enterprise administrators can enable a beta. Pro access initially excludes the EEA, Switzerland and the UK. Eligibility does not guarantee an immediate rollout.
The first dot comes with eligible plans without an additional creation fee, though deeper work has usage limits. Set one up on desktop or desktop web; afterwards you can chat through a supported mobile app.
Why does it feel like it understands the job?
Powered by GPT-6 Astra, a dot has its own cloud computer and can use memory, connected apps and scheduled tasks to keep projects moving between conversations.
Think of a research assignment followed by a finished draft and another round of edits. An assistant that remembers the context doesn't need the entire backstory every time. That alone can feel impressively intelligent.
It doesn't read minds. Permissions, approvals, task-specific context and reviewing the result still matter. What changes is the emphasis on finishing ongoing work, not merely answering the next message.
The moment Codex becomes frustrating
One reported experience involved asking an agent to revise and save a document. Instead, it kept collecting possible problems, requesting clarification and consuming time and allowance without leaving a finished file.
Quality checks matter. An inspection meeting is not a deliverable.
That's an anecdote about a particular workflow, not a scientific measurement of all Codex users. Different models, tasks, permissions and environments can lead to different results.
Plot twist: Dots can delegate to Codex
OpenAI describes Codex as supporting app interaction, memory and ongoing automation as well as coding. Editing software, running tests and reviewing changes remain major strengths.
And the official dots guide says a dot can start tasks in a Codex cloud environment you've already configured. They don't have to be rivals: Dots can coordinate the long-running objective while Codex handles technical implementation.
Two AIs dividing the work. Maybe the human finally gets to retire from chasing status updates.
Count deliverables, not intelligence trophies
Give both tools comparable tasks and measure completed files verified after saving, how often instructions had to be repeated, and the time plus usage required to finish.
A brilliant answer without the requested file is still an unfinished job. A quieter assistant that understands the aim and completes it may be more useful.
It's reasonable for Dots to feel smarter in your workflow. It does not, by itself, prove a universal reasoning advantage over Codex.
The best AI response might simply be: “Done. Here's the finished work.”
