In plain English
An API lets one program send requests to a service and get answers back, usually billed per use. A decision model like Luna or Jev picks one answer from a fixed list of possible answers. Codex is OpenAI's coding agent.
OpenAI made a pile of announcements at DevDay on 29 September. We'll see as things unfold.
The Decisions API looks good, especially on the heels of Jev. OpenAI says it "enables real-time decision-making by focusing Luna's intelligence on a specific set of user-defined questions with finite pre-defined answers."
I have to test the Agents API: hosted agents with memory and computer use in an OpenAI-hosted browser. OpenAI says GPT-6.1 Sol "delivers near-Astra intelligence to everyone at a fifth of its standard input and output token prices"; I haven't tested it. Codex running cloud tasks with the laptop closed is good, I've tried the CLI's two-way voice and it's quite good, and I still have to try the security scanning. ChatGPT Space looked good for people working with agents.
Dots sounds clever. I'm not sure how long it'll be around; it sounds like a cute thing. OpenAI is rolling it out "in eligible markets," and I'm in the EU, so I won't be using it.
Pro 500 set the cat amongst the pigeons. It fragments the offering further, and people who resubscribed to the 200 tier got a little bit burned on usage. From 30 October, Pro 200 drops from 20 to 10 times the Plus allowance in ChatGPT Work and Codex, according to The Next Web. The token reset was a bit of a gimmick. More running in the cloud on subscription plans rather than API keys is better.