OpenAI adds GPT-6 Astra, Sol and Luna to ChatGPT's Work and Codex tabs, while GPT-6 Pro reaches Chat on higher-tier plans

The Chat/Work control is a slider at the top of the ChatGPT window, and TechRadar's reviewer said it was the one thing on screen they never bothered with, because Chat did almost everything they wanted ^[13][12]. Anyone who told a team "use ChatGPT" last year was building on that habit. People open the app and type into the prompt bar in front of them.

Work uses the same prompt bar, in the same place ^[2]. Behind it sits a different model list. TechRadar describes Work as handing over more of the job: a larger multi-step task where ChatGPT researches, reads files or uses websites instead of waiting for direction at every stage ^[14]. Replies take considerably longer as a result, and the reviewer said typing ordinary questions into Work is "not a particularly good experience" ^[3][4].

That cuts both ways in a rollout. Someone who wanders into Work for a quick answer waits, dislikes it, and goes back to Chat. Someone with a five-file job sits in Chat steering it by hand and never learns the other tab exists.

On the paid-surface question, the evidence lands slightly off where the tab boundary is. The reviewer holds ChatGPT Plus and reaches Astra, Sol and Luna through Work and Codex ^[9]. Work is a tab the Plus subscription already includes. What money buys is Astra inside Chat, by way of GPT-6 Pro on some higher-tier plans ^[8]. TechRadar does not name those plans or say what they cost ^[15].

The picker is the part an enablement doc has to survive. Five entries show up in Work on that Plus account: two Astra sizes, two Sol sizes, and Luna at High ^[10][11]. In Chat, the same reviewer had been working in GPT-5.6 Sol ^[12]. OpenAI says the smaller models "build on the advances behind GPT-6 Astra" and bring much of its strengths into faster and cheaper form ^[7]. That is useful framing for a pricing page. It is no help at all to a person choosing between Sol Light and Sol Medium at 9am.

Take the five things your people actually do in ChatGPT most often, and write two columns next to each: the tab, and the exact picker entry. A question with an answer stays in Chat, and the instruction can stay as vague as it is now. A job with several files or a browse step belongs in Work, and the instruction has to name the tab before it names the model, because the interface gives no cue. A task where someone insists on Astra in the Chat window is a billing question, and it goes to whoever owns the plan.

If you cannot fill in those two columns for all five rows, the doc on your intranet is now wrong, and the people following it are on last month's model. OpenAI's own pitch for Astra is operating software, browsing, research and multistep professional workflows ^[6]. Those are Work-column tasks.