How to Use GPT-6 Astra: The Complete Access and Setup Guide (2026)

GPT-6 Astra is OpenAI’s newest flagship model, released on September 3, 2026. Using it is mostly a question of access: depending on your plan, Astra appears under a different name in the ChatGPT model picker, lives in Work or Codex, or comes through the API. This guide covers every route, its cost, and what to do when Astra is missing.

That access question is the whole story right now. Astra is the first OpenAI model to hit the company’s “Critical” cybersecurity threshold, so OpenAI is letting accounts in in stages — and eighteen days after launch, most people still can’t find it. Below: the rollout order, where the model hides in each app, what the API costs, and how to get good results once you’re in. Think of this page as the definitive walkthrough for how to use GPT-6 Astra across every surface, so you can skip the guesswork.

If you missed the launch itself — what Astra does and why its security rating matters — our explainer covers it. This one is the practical companion: nothing but getting in and getting work done.

How to Use GPT-6 Astra: Start With Access, Not the App

The first thing to understand about how to use GPT-6 Astra is that the bottleneck is rarely the interface — it’s eligibility. Astra is the first OpenAI model to reach the company’s “Critical” cybersecurity threshold under its Preparedness Framework, a classification reserved for capabilities that could pose serious risks if misused. Rather than opening every capability to the public at once, OpenAI is taking a phased approach to the release. (For the full picture of what Astra is, see our GPT-6 Astra explainer.)

Per OpenAI’s announcement: a limited set of organizations first — including cybersecurity customers in the company’s Daybreak program — then Plus, Pro, Business, and Enterprise users, plus the API and AWS. Free-tier users and the cheapest paid plan are excluded in the near term.

The practical upshot: whether you can use Astra today comes down to your plan and whether your account has been let in yet. If you opened ChatGPT expecting a button labeled “Astra” and found nothing, that’s the normal experience for most accounts right now — not something you broke.

Find GPT-6 Astra in the ChatGPT Model Picker

Assuming your plan qualifies, the next hurdle is simply finding the model — and this is where most people stall, because in a regular ChatGPT conversation the model doesn’t call itself Astra at all. Follow the steps below and you’ll know exactly how to use GPT-6 Astra in regular Chat, whichever plan you’re on.

1. Sign in with an eligible plan. You need Plus, Pro, Business, or Enterprise. Free and Go accounts are excluded from this rollout for now.
2. In regular Chat, look for “GPT-6 Pro.” On the $100 and $200 Pro tiers, plus Business and Enterprise, the model picker’s Pro section lists an entry called GPT-6 Pro — that is the one powered by GPT-6 Astra. Don’t keep hunting the menu for the flashier name; you won’t find it there.
3. On Plus, switch surfaces instead. Plus subscribers get Astra through ChatGPT Work and Codex, not regular Chat — open Work on web or mobile and choose Astra there, or in the desktop app switch from Chat to Work first.
4. Update Codex before expecting it. Codex users should make sure the Codex CLI is at version 0.153.0 or newer — older versions won’t list Astra even on an eligible account.
5. Enterprise users: check with your admin. Access can also depend on workspace model-access permissions, and OpenAI ships Astra switched off by default for Enterprise workspaces — an administrator has to enable it.
6. Update the app and allow for the stagger. Even if the desktop app recently updated itself, OpenAI suggests a manual check (Menu > Check for Updates) — Work rolls out separately, so Astra can be live on one platform before another.

Here’s the whole picture in one place:

Plan	Astra access	Where to find it
Free / Go	Not available	Excluded from the rollout for now
ChatGPT Plus	Yes, as the rollout reaches your account	ChatGPT Work or Codex — not regular Chat
ChatGPT Pro ($100 / $200)	Yes	“GPT-6 Pro” in the Chat model picker, plus Work and Codex
Business	Yes	“GPT-6 Pro” in the Chat model picker, plus Work and Codex
Enterprise	Yes, once enabled	Same as Business — but an admin must switch it on; it’s off by default

Use GPT-6 Astra in ChatGPT Work and Codex

ChatGPT Work is the part of this story most people underestimate. Work is the mode where Astra operates as an agent rather than a chatbot — reading files, driving applications, carrying out multi-step tasks — and for Plus subscribers it is currently the only way in. If you’re on Plus and staring at a Chat window wondering where Astra is, you are simply in the wrong room. Learning how to use GPT-6 Astra inside Work is therefore the fastest path for most subscribers.

Astra’s agent side is the point of the release. OpenAI describes the model as able to work through screens and controls instead of needing an API integration for every application — the pitch being that it finishes the job rather than telling you how to do it. In Codex, long sessions get a new trick: instead of compressing everything into one summary each time the context window fills, Astra can keep notes across context windows, with earlier windows remaining searchable. That feature is experimental — you switch it on in your Codex config — and OpenAI says it will become the default for Astra in the coming weeks.

Two practical notes. First, privacy: even when Work runs locally with your files, messages and task context may still be stored in the cloud — don’t hand it anything you wouldn’t want on OpenAI’s servers. Second, billing: Astra usage is included in existing subscription allowances, with credits available for extra usage; Pro, Business, and Enterprise also get GPT-6 Astra Pro for the longest tasks, with separate limits.

Set Up the GPT-6 Astra API Without Blowing Your Budget

For developers, the API is the most direct route to Astra — and the one where sloppy defaults get expensive fast. Here’s a setup sequence that keeps the first invoice boring. Follow it and you’ll know how to use GPT-6 Astra through the API without a surprise bill.

1. Use the Responses API, not Chat Completions. Astra works on both, but reasoning state, compaction, and hosted tools are Responses features — anything agentic should go through Responses.
2. Start your reasoning effort at medium. Run every new task class at medium first and climb only if it actually fails — the top of the ladder is where the money goes.
3. Draw a hard line at 272,000 input tokens. Requests above that threshold bill at 2× input and 1.5× output — for the entire request, not just the overflow. Design your retrieval strategy so no single request crosses it.
4. Lean on the cache. Prompt caching brings repeated input down to about $1 per million tokens, which changes the math completely for agentic loops that re-read the same context every turn.
5. Put a budget guard on every call. Check response usage and alert on outliers — at $50 per million output tokens, a runaway loop becomes a surprise invoice fast.
6. Run one read-only task end to end first. Give Astra a job that produces a written deliverable and touches nothing — then grant write access.

The headline numbers: $10 per million input tokens and $50 per million output tokens, with a context window of roughly 1.05 million tokens. For scale — a request with 100,000 uncached input tokens and 10,000 output tokens costs about $1.50. Cached reads are far cheaper, and output discipline matters: standard output tokens cost five times what input does.

And two hard limits worth knowing up front: fine-tuning is not supported on Astra, and there are no open weights — it’s a closed, hosted model, so self-hosting is not on the table.

Troubleshooting: When GPT-6 Astra Doesn’t Show Up

Most “missing Astra” cases trace back to a short list of causes. Check these before assuming anything is broken. Run through this list whenever you’re stuck on how to use GPT-6 Astra on a new device or workspace.

• It’s simply not your turn yet. The rollout is gradual and OpenAI hasn’t published an activation schedule. If your plan qualifies and everything below checks out, waiting is the fix.
• You’re looking for the wrong name. In regular Chat the entry is “GPT-6 Pro,” not “Astra.” If you see GPT-6 Pro, you already have it.
• You’re in Chat when you should be in Work. Plus subscribers only get Astra through Work and Codex — it will never appear in regular Chat.
• Your app is behind. Manually check for updates (Menu > Check for Updates) — the auto-updater can lag the rollout.
• Your workspace is blocking it. On Enterprise, Astra is off by default; an admin has to enable it, and workspace model-access permissions can hide it even after that.
• Your Codex CLI is old. Below version 0.153.0, Astra won’t appear in Codex regardless of plan.
• The API says the model doesn’t exist. Confirm you’re calling the gpt-6-astra model id and that your account’s rollout status actually includes API access — eligibility differs between ChatGPT and API surfaces.

Prompting Astra Well: Five Habits That Pay Off

Once you’re in, Astra rewards a different style than older models — less chatting, more delegating. Five habits keep showing up in what experienced users report. Master these and you’ll genuinely know how to use GPT-6 Astra day to day, not just in a demo.

1. Give it a finish line. “Help me analyze this” leaves too many gaps — state what the session should produce: a ranked list, a spreadsheet, a finished deck. Astra asks for clarification more than older models; feed it the decisions it actually needs.
2. Define the assignment’s boundaries. Spell out goal, allowed actions, forbidden actions, and success criteria. For agentic work, explicit permissions aren’t plumbing — they’re part of the design.
3. Tell it what each file is for. When you attach or reference files, add one line per file — evidence, template, style reference. That single habit kills most of the back-and-forth clarifying rounds.
4. Start read-only, add checkpoints. Begin with a research or planning pass, review the evidence, then authorize the smallest set of changes that gets the job done. For anything consequential — deployments, purchases, account changes, destructive commands — require a confirmation stop before it proceeds.
5. Interrupt it mid-task. You don’t have to wait for a run to finish or start over. Astra picks up a new instruction mid-stream and updates its plan from there — useful when you spot it heading down the wrong path early.

What This Means Practically

Stepping back, the Astra rollout previews how frontier model releases will work from here: the capability lands first, and access, pricing, and safety gating decide who actually gets to use it. The “Critical” cybersecurity rating is the reason for the velvet rope — and worth taking seriously rather than treating the wait as a marketing stunt. OpenAI’s own safety researchers have documented real misalignment cases in advanced models, as we covered in our breakdown of OpenAI’s AI safety cases and disclosure framework; staged access to a model this capable is the cautious move.

For individuals, the decision tree is simple. On Free or Go: there’s nothing to do but wait — no extension, activation code, or unofficial site will unlock Astra, and anyone claiming otherwise is selling something. On Plus: your path is Work and Codex, so get comfortable with the Work mode now instead of hunting the Chat picker. On Pro, Business, or Enterprise: look for GPT-6 Pro in the picker, sort out admin permissions early on a team, and remember Astra Pro exists for the longest tasks.

The short version of how to use GPT-6 Astra: confirm your eligibility, pick the right surface for your plan, and keep an eye on costs.

For developers, the API math is the real story. At $10/$50 per million tokens with a million-token context window, Astra is priced for serious work — which is why the medium-first effort habit and the 272,000-token line matter. Get those two right and the model is affordable; ignore them and the invoice teaches the lesson instead.

FAQ

When will I get GPT-6 Astra on my account?

OpenAI hasn’t published a schedule. The rollout began September 3, 2026 with a limited set of organizations and is expanding to Plus, Pro, Business, and Enterprise users, plus the API, in stages. Being on an eligible plan doesn’t mean it’s available to you today — availability also depends on your client, sign-in method, and workspace settings.

Can I use GPT-6 Astra with ChatGPT Plus?

Yes, once the rollout reaches your account — but only through ChatGPT Work and Codex, not regular Chat. In the desktop app, switch from Chat to Work and select Astra; on web or mobile, open Work directly.

Why does my model picker show “GPT-6 Pro” instead of Astra?

That’s Astra. In regular ChatGPT conversations on Pro, Business, and Enterprise plans, the model is listed as “GPT-6 Pro” — the entry powered by GPT-6 Astra. If you see it, you already have access.

Is GPT-6 Astra free?

No. Free-tier users and subscribers on OpenAI’s cheapest paid plan are excluded from the rollout in the near term. For paid plans, Astra usage is included within existing subscription allowances, with the option to purchase credits for additional usage.

How much does the GPT-6 Astra API cost?

$10 per million input tokens and $50 per million output tokens, with cached input reads around $1 per million. Requests above 272,000 input tokens are billed at 2× input and 1.5× output for the entire request. As a reference point, 100,000 uncached input tokens plus 10,000 output tokens costs about $1.50.

Does GPT-6 Astra support fine-tuning?

No. Fine-tuning is not supported, and there are no open weights — Astra is a closed, hosted model, so self-hosting isn’t possible either.

References

1. OpenAI — “GPT-6 Astra” (official announcement), https://openai.com/index/gpt-6-astra/ — September 3, 2026.
2. BetaNews — “OpenAI launches GPT-6 Astra, claims AGI era has begun”, https://betanews.com/article/openai-gpt-6-astra-agi-era/ — September 3, 2026.
3. 9to5Mac — “OpenAI releasing major upgrade to ChatGPT and Codex with GPT-6 Astra, details here”, http://9to5mac.com/2026/09/04/openai-releasing-major-upgrade-to-chatgpt-and-codex-with-gpt-6-astra-details-here/ — September 4, 2026.
4. Engadget — “How to use GPT-6 Astra when it rolls out to you”, https://www.engadget.com/2252859/how-to-use-gpt-6-astra-rollout-schedule/ — September 14, 2026.
5. Tech Startups — “OpenAI launches GPT-6 Astra, its most powerful AI model with advanced cybersecurity capabilities”, https://techstartups.com/2026/09/04/openai-launches-gpt-6-astra-its-most-powerful-ai-model-with-advanced-cybersecurity-capabilities/ — September 4, 2026.
6. DataNorth — “OpenAI GPT-6 Astra: Price, Benchmarks and Access”, https://datanorth.ai/news/openai-launches-gpt-6-astra — September 4, 2026.
7. AIstify — “OpenAI Launches GPT-6 Astra, Its Most Powerful Model Yet”, https://aistify.com/openai-gpt-6-astra-computer-use-model/ — September 4, 2026.
8. Sabbirz — “GPT-6 Astra Complete Guide”, https://sabbirz.com/blog/gpt-6-astra-complete-guide — September 4, 2026.
9. DEV Community — “GPT-6 Astra Guide: How to Use It, Best Prompts, Pricing & Use Cases (2026)”, https://dev.to/maestro_morty_cc62e2c85ae/gpt-6-astra-guide-how-to-use-it-best-prompts-pricing-use-cases-2026-dlh — September 4, 2026.
10. Tech-Insider — “How to Use GPT-6 Astra API: 12 Steps [2026]”, https://tech-insider.org/how-to-use-gpt-6-astra-api-2026/ — September 2026.
11. Eric Talks (Substack) — “How to Upgrade ChatGPT to GPT-6 Astra: Access, Plans, Model Picker, and Rollout Guide”, https://erictalks.substack.com/p/how-to-upgrade-chatgpt-to-gpt-6-astra-2026/ — September 4, 2026.
12. TechBullion — “How to Use GPT-6 Astra to Create Research-Oriented Slide Decks”, https://techbullion.com/how-to-use-gpt-6-astra-to-create-research-oriented-slide-decks/ — September 2026.