What Is GPT-6 Astra? A Job Seeker's Guide (vs Sol and Luna) + 8 Prompts

What GPT-6 Astra is, in one paragraph

GPT-6 Astra is the top model in OpenAI's GPT-6 family. It launched on September 3, 2026 after OpenAI delayed it to add safeguards, and it became the default engine in ChatGPT's paid plans on September 25. OpenAI built it on its largest training run to date and pitches it at long, multi-step professional work: using a computer, browsing, writing, coding and research. OpenAI's own list of example tasks includes filing tax returns, ordering food and running a job search.

It's also the model underneath OpenAI's new dots, the always-on personal agents OpenAI launched at DevDay on September 29. If you've seen headlines about AI agents applying to jobs for people, Astra is the engine most of those stories are about.

For you, the practical question is simpler: what does Astra do better than the ChatGPT you were using in August, and when is it overkill?

Astra vs Sol vs Luna: which one are you using?

OpenAI now has four GPT-6 models, and ChatGPT picks between them depending on your plan and which mode you're in. Here's the job-seeker version of the family tree.

GPT-6 Astra is the flagship: the most capable, the slowest and the most expensive through the API. It's the default in paid ChatGPT.

GPT-6 Sol (September 22) is the everyday workhorse, priced at $2 per million input tokens and $10 per million output. OpenAI says it makes about half as many mistakes as GPT-5.6 Sol. GPT-6.1 Sol, announced at DevDay on September 29, is an upgrade that OpenAI says gets close to Astra on professional work at roughly one-fifth of Astra's token price.

GPT-6 Luna (September 22) is the fast, cheap model for simple tasks. Free and Go users get it in the ChatGPT desktop app.

What this means in practice: if you're on Plus or higher, you're already using Astra in normal chats. If you're on the free plan, you're using Luna, which is fine for formatting and short rewrites and noticeably weaker for anything that needs judgment.

Where Astra actually helps a job search

Reading many job ads at once. Astra's large context window means you can paste ten postings for similar roles and ask what they have in common. That's the fastest way to find the skills a market actually wants, instead of guessing from one ad.

Company research before an interview. Astra is strong at browsing and synthesising. Ask it to pull a company's last two earnings calls, recent product launches and leadership changes, then tell you the three problems the team you're interviewing with is probably trying to solve. That's better interview prep than anything a generic question list gives you.

Keeping a structured search on track. Give Astra your target roles, your constraints (location, salary floor, remote or not) and a list of applications, and it's good at keeping that state straight over a long conversation: what you've sent, what's pending, who to follow up with.

Understanding what a vague ad really wants. Astra is better than earlier models at reading between the lines. An ad that mentions 'ambiguity' three times is telling you something, and Astra is more likely to spot it and suggest which story from your experience to lead with.

Where Astra still slips

It invents numbers. Give Astra a vague bullet like 'improved onboarding' and it will happily return 'reduced onboarding time by 30%'. It sounds right. It isn't yours. Always give it a fact sheet of your real metrics and tell it to mark any bullet where it wanted to add a number it didn't have.

Its writing is recognisable. Astra inherits the ChatGPT house style: smooth, confident, well-structured. Recruiters have read a thousand CVs and cover letters in that voice since 2023. Give it a sample of your own writing and tell it to match that.

It doesn't know what an ATS will do with your file. Astra writes text. Whether Workday, Greenhouse or Taleo can parse the PDF you export is a separate question it can't answer, even when it sounds sure.

It's overkill for small jobs. Using Astra to fix a typo or reformat a skills section is like hiring a consultant to move a chair. If you're on the API, use Sol or Luna for those.

8 prompts to use GPT-6 Astra on your job search

1 — Market scan: "Here are 10 job ads for [role]. List the skills and tools that appear in at least 6 of them, the ones that appear in 3–5, and any that appear only once. Quote the ad text for each."

2 — Gap check: "Here is my CV. Against the 'at least 6' list above, tell me which skills my CV shows clearly, weakly or not at all. Cite the line from my CV."

3 — Honest bullet rewrite: "Rewrite my experience bullets for [target job]. Use only numbers from this fact sheet. Mark any bullet where you wanted to add a metric with [NEEDS NUMBER]. No 'passionate', no 'results-driven'."

4 — Company brief: "Research [company]. Summarise recent product launches, leadership changes and anything the team for [role] is likely struggling with. Link every source."

5 — Interview angle: "Based on that brief and my CV, which two stories from my experience should I lead with in the interview, and why?"

6 — Search tracker: "Here are the applications I've sent this week with dates and status. Keep track. When I paste an update, tell me who to follow up with and when."

7 — Honesty audit: "Compare the CV you wrote with my original CV and fact sheet. List every claim, number or skill in your version that I didn't give you."

8 — Plain-text export: "Output the final CV as plain text: no tables, columns, icons or text boxes. Use standard section headings."

Should you let an Astra agent apply for you?

OpenAI's dots, launched September 29, are always-on agents powered by Astra with their own cloud computer. In principle, one could search for roles, tailor a CV and submit applications while you sleep. They're rolling out first to ChatGPT Pro and Business Premium users in eligible markets.

Our advice: use an agent to do the research and prepare drafts, and keep the 'submit' button for yourself. Mass-applying with near-identical tailored CVs is easy for recruiters to spot, and one wrong claim sent to fifty employers is fifty chances to be caught. Also think carefully before giving any agent the logins to your job-board accounts.

Whatever does the writing, the file still has to survive screening.

The step Astra can't do: test the file

When your Astra-tailored CV is ready, export it to PDF or Word and run it through a free analysis at https://www.hrlens.io/cv. You get a score out of 100, five category scores (experience, tech stack, impact and ownership, clarity and structure, ATS compatibility), your strengths and a visual layout check.

Paste the same job description into the job-targeting field and the analysis lists the skills the ad asks for that your CV still doesn't show, then generates ATS-friendly rewritten versions in six templates that mirror the job's keywords. Paste the ad text rather than a LinkedIn or Indeed link; those sites block fetching.

If the score comes back low, feed the missing-skills list back into prompt 2 and go round again. Astra does the thinking; a real parser confirms a machine can read the result.

Frequently asked questions