How to Stop Running Out of ChatGPT-6 Astra Usage: 2 Settings and 5 Prompts

You paid for ChatGPT-6 Astra, and it can run a whole job for you: sort the inbox, build the spreadsheet, research the purchase, prep the week. Then halfway through the afternoon it tells you that you’ve hit your limit and to try again in a few hours. That’s fixable. Two settings and a better way of asking get you far more out of the same $20 plan, and you’ll stop watching the meter.

ChatGPT-5.6 Sol answered questions. You typed, it replied, one message, done. ChatGPT-6 Astra does jobs. It opens a browser, reads pages, clicks, builds documents, checks its own work. Every step it takes, and every question it stops to ask you, counts against the same five-hour allowance. That’s why the same $20 plan feels smaller with Astra: you’re paying per job, and a job can be one message or ten.

Three things turn one job into ten messages:

The fix for the first one is a setting. The fix for the other two is how you write the prompt, and on the desktop app, Goal mode.

Medium. Unless you’re on the $200 Pro plan, keep ChatGPT-6 Astra on medium and leave it there. Simple prompt, complex prompt, doesn’t matter.

Here’s why that isn’t a compromise. On September 6, an OpenAI staff member wrote in OpenAI’s developer forum that GPT-6 Astra on low already performs better than GPT-5.6 Sol on high, and told people who were happy with Sol on high to move down to low or medium on Astra. Medium is already a step above the best setting you had before. Independent benchmarks published the same week found that going from medium to high on Astra bought about one point on their scale for about 22% more cost. You would never notice the point. You will notice the usage.

Where the setting is. Tap the model name at the top of the chat. The reasoning level sits with it, and the same five labels show up on every model: Light, Medium, High, Extra High, Max. The advice in this guide is for ChatGPT-6 Astra in Work, because that’s where the allowance is small and the jobs are long. Switch to Work, check that the level says Medium, and leave it there.

Goal mode keeps ChatGPT-6 Astra pointed at one finish line until it gets there. Instead of doing a step and handing the conversation back to you, Astra keeps working, checks its own progress against the goal you gave it, and stops when the goal is met. That’s what stops the wandering, and wandering is what burns your usage.

How to start it. In the ChatGPT desktop app, type /goal in the message box, then describe the goal. Three things belong in that description, and they’re the same three parts as any good Astra prompt: the outcome you want, the constraints, and what a finished result looks like. Goal mode needs that last one most. If you can’t say what done looks like, Astra can’t tell when to stop.

While it runs. A progress row appears above the message box. From there you can pause, resume, edit the goal, or clear it. You can also keep typing: send a follow-up to add a detail or tighten a constraint, and Astra folds it in without starting over. If you want to know where it’s at, ask it to summarize progress. Keep related work in the same chat so it doesn’t lose the thread.

On the phone or in a browser. There’s no /goal there. Use ChatGPT Work and put the outcome, the constraints, and the finished result directly in your prompt. You get most of the benefit, because the reason Goal mode works is that it forces those three things to be said. Prompt 5 below is written that way.

When to use it. Goal mode is for jobs with several steps and a clear end: build the whole spreadsheet, research all fifteen, get the document to final. For a two-minute job, a normal prompt is enough.

Three parts, in order. Start with an action verb: open, search, read, build. Give it constraints: what counts, what to skip, what not to touch, and how to handle a question without asking you. End with what done looks like: the exact shape you want back. The full formula, with ten prompts for everyday jobs, is in How to Prompt ChatGPT-6 Astra. The five below add one thing that guide doesn’t: each one is built to cost one message.

The line that does most of the saving is this one, and it’s in every prompt: “Don’t stop to ask me questions. If something is unclear, make a sensible choice, keep going, and list the choices you made at the end.” That single sentence turns a four-message job into a one-message job.

About the brackets. Anything in [square brackets] is a blank for you to fill in before you paste. Type your real detail in its place and delete the brackets. When Astra reaches a page that needs a login, it shows you a sign-in form. Type your password into that form, not into the chat. The form sends it straight to the browser, and the model never sees it.

1. The one-run inbox

Sort, unsubscribe, and draft the replies, all in one pass. Most people run this as three separate prompts. It’s one.

Open my email and go through everything from the last 7 days. Unsubscribe from newsletters that are sitting unread from the last 30 days. Archive receipts, shipping notices, and automatic notifications. Leave everything else where it is. Don't delete anything and don't send anything. For each message that needs a reply from me, write a two-sentence draft in a plain, friendly tone. Don't stop to ask me questions. If something is unclear, make a sensible choice, keep going, and list the choices you made at the end. Done looks like: one list with the sender, the subject, and the draft reply for each one, followed by a count of what you unsubscribed and archived.

2. The buy-or-don’t brief

Everything you’d spend an evening reading, in one table with a pick at the bottom.

Research [the product or service you're deciding on] for [what you'll use it for]. Compare the 3 best options for someone who [your situation: a small business, a family of four, a beginner]. Use the companies' own sites and two independent reviews each. Ignore sponsored lists. Don't add anything to a cart, sign up for anything, or contact anyone. Don't stop to ask me questions. If something is unclear, make a sensible choice, keep going, and list the choices you made at the end. Done looks like: a table with price, what's included, the best thing about each, the one drawback, and the return policy, then your pick in two sentences.

3. The long document, decided

Attach a contract, a policy, a report, or a proposal, and get back the version you can act on.

Read the attached document all the way through. Write a one-page plain-English summary for someone who hasn't read it. Then list every date, deadline, fee, and obligation it puts on me, with the page or section it comes from. Then list the 5 questions I should ask before I sign or agree. Don't summarize anything you didn't read, and don't guess at numbers. Don't stop to ask me questions. If something is unclear, make a sensible choice, keep going, and list the choices you made at the end. Done looks like: those three sections, in that order, on one page.

4. The application pack

A posting and your resume go in. A tailored resume, a cover letter, and an honest gap list come out, in one run.

Open this job posting: [link]. Read the attached resume. Rewrite my resume bullets to match the posting's wording where it's true, and don't invent anything I didn't do. Then write a 150-word cover letter in a plain, confident voice that names the company and the role. Then list every requirement in the posting that my resume doesn't show, so I can decide whether to address it. Don't apply, don't submit anything, and don't message anyone. Don't stop to ask me questions. If something is unclear, make a sensible choice, keep going, and list the choices you made at the end. Done looks like: the revised resume, the cover letter, and the gap list.

5. The Goal-mode job

A real multi-step job with a finish line built in. On the desktop app, type /goal and paste it. On the phone, paste it into Work as is.

Build a spreadsheet of 15 [type of business or vendor] within 20 miles of [city]. Use each one's own website and their Google Maps listing. For each, get the name, the address, the phone number, the website, their hours, their starting price or price range if it's listed, and their Google rating. Skip any that have been closed or have no website. Don't fill out forms, book anything, or contact anyone. Don't stop to ask me questions. If something is unclear, make a sensible choice, keep going, and list the choices you made at the end. Done looks like: a spreadsheet with those 8 columns and exactly 15 rows, sorted by rating, highest first. Stop when there are 15 rows.

When a job is eating usage, what comes back tells you which fix you skipped.

The limit is still the limit. Medium and one-run prompts stretch the allowance; they don’t remove it. The five-hour window resets on its own, and there’s a weekly cap on top. ChatGPT tells you when you’ve hit the window and when it opens again. The honest way to plan is to save Astra for jobs that take clicking and give quick questions to regular chat, which has its own limits.

Codex comes out of the same pool. If you also use Codex, its jobs and your Work jobs share one allowance, so a heavy afternoon in one leaves less for the other.

You can buy a reset, in some countries. On eligible personal Plus and Pro accounts, OpenAI sells an instant reset that refreshes the five-hour and weekly allowance the moment checkout goes through. It’s there if you need it on a deadline. The two settings above mean you mostly won’t.

• OpenAI Help Center, “Managing usage with GPT-6 Astra in Work and Codex”: the 5 to 45 messages per five-hour window on Plus, the weekly window, the effort guidance, and purchased resets. help.openai.com
• ChatGPT Learn, “Long-running work”: the /goal command, the progress row, and the web guidance. learn.chatgpt.com
• ChatGPT Learn, “What’s new,” August 3 to 7, 2026: Work and Codex share the same usage limits and credits. learn.chatgpt.com
• OpenAI Developer Community, September 6 to 7, 2026: OpenAI staff on Astra low versus Sol high. community.openai.com