How to Use GPT-6 Astra (and Is It Free?)

Our GPT-6 Astra launch write-up covered what the model is, the benchmark table and the price tag. Two days later the question landing in the inbox is a different one: “I pay for Plus, there is no Astra anywhere in my model picker, where is it?”

The answer is simple but not intuitive. OpenAI said the model would reach “all ChatGPT Plus, Pro, Business and Enterprise users”, then put the Plus-tier access not in the ordinary chat window but inside ChatGPT’s Work and Codex experiences. Your subscription is fine; you are looking in the wrong place.

Below: where to open Astra on each plan, how many messages you actually get, how to use it in Codex and the API, and how to keep the bill under control. 👇🏻

Which Plans Actually Have Astra?

ChatGPT is no longer a single chat box. There are three experiences, and Astra is not landing in all of them at once:

• Chat: quick questions, everyday conversation, search.
• Work: an agent for longer, multi-step jobs that produce finished deliverables. Reports, decks, spreadsheets, documents, Sites.
• Codex: software development and technical work. Local folders, repositories, terminals.

Astra by plan:

The Plus row is the one that trips people up. The Pro tier in the model picker (the GPT-6 Pro entry) is not offered on Plus; a Plus subscriber reaches Astra through Work and Codex.

Because the rollout is gradual, two accounts on the same plan can differ: one sees the model, the other does not. That is queue position, not a bug.

Step by Step in ChatGPT

On Pro, Business or Enterprise

1. Open ChatGPT and switch to the workspace you intend to use.
2. Open the model picker above the composer.
3. Below the speed and reasoning tiers (Instant, Medium, High, Extra High), open the Pro option.
4. If GPT-6 Pro is listed, Astra is live on your account. Select it and prompt as usual.

Enterprise has one extra step: access ships off by default. Until a workspace admin enables the model permission in Workspace settings, the entry never appears.

On Plus (through Work)

1. Update and open the ChatGPT desktop app.
2. Select ChatGPT from the top-left menu.
3. Use the Chat / Work toggle at the top of the page and switch to Work.
4. Start a new Work chat and describe the deliverable you want.

Work also exists on web and mobile, where it runs in the cloud. In the desktop app, and only where your plan and workspace allow it, Work can reach your local files and desktop apps with your permission. That is the real difference: Work does not just answer, it finishes the job.

Work itself is rolling out gradually. If it is not in your top-left menu yet, your turn has not come.

Voice

ChatGPT Voice now works inside Work and Codex as well (macOS and Windows desktop app). Select the Voice control, grant microphone access, and you can start tasks by speaking and interrupt the model mid-sentence. If it needs computer context, screen recording and accessibility permissions are requested too. Only one Voice conversation runs at a time.

“Astra Is Not Showing Up” Checklist

Work through these five, in order. They cover almost every case:

1. Is your plan eligible? Free and Go never show Astra anywhere.
2. Are you in the right surface? On Plus, stop looking in Chat; check Work and Codex.
3. Are your clients current? Astra needs Codex CLI 0.153.0 or newer, plus the latest desktop app.
4. Is admin permission granted? Business and Enterprise admins can restrict model access, and on Enterprise the default is off.
5. Has the rollout reached you? It moves at different speeds across Chat, Work and Codex. Buying credits does not grant early access.

Country availability is not a separate list: Astra follows ChatGPT’s supported countries.

Message Limits: “I Pay for Pro, So It’s Unlimited?”

No. A Pro subscription does not include unlimited GPT-6 Pro. In Chat, the allowances are:

Work and Codex allowances are counted separately. Pro $100, Pro $200 and Business Premium seats can spend their full existing Work/Codex allowance on Astra. Plus and Business Standard seats get limited Astra usage, with optional credits afterwards.

On Pro $200, hitting the weekly GPT-6 Pro limit drops you to GPT-5.6 Thinking at Medium. On Pro $100 both models draw from one pool, so switching between them buys you nothing.

Is GPT-6 Astra Free?

Short answer: no, there is no free Astra access.

• Free and Go default to GPT-5.6 Luna. The Think button exists for harder questions, but it also runs Luna, not Sol and not Astra.
• Plus ($20/month) unlocks Astra inside Work and Codex, in limited quantity.
• Astra usage in ChatGPT counts against your existing subscription allowance; once it runs out you can buy credits for more.
• The API has no free tier. You pay per token.

So the honest answer to “how do I use GPT-6 Astra for free” today is: you cannot. The cheapest entry point is the limited allowance inside a Plus subscription’s Work experience.

What Can You Actually Get Astra to Do?

Astra’s real separation is not chat quality, it is computer use. OpenAI’s own examples are jobs the model finishes by looking at a screen and driving the keyboard and mouse:

• Electronics: laying out a PCB in KiCad, taking a schematic and placing components and routing copper. If you work with Arduino or embedded hardware, this is the most interesting item on the list.
• 3D and games: modelling a house in Blender and turning it into a walkable Unreal Engine 5 scene, or building a kart racing game from scratch.
• Office work: producing slides that respect your template, building spreadsheets, formatting a legal contract, working inside Power BI.
• Ordinary drudgery: filling out forms, updating CRM records, organising a calendar, running online research and drafting the summary straight into your email or document editor.
• Web: building a site and running its front-end QA checks itself. With Sites in ChatGPT it can create, host and share sites, web apps and games from a single prompt.

The speed gain is concrete too. In OSWorld 2.0 latency simulations, Astra scores 72.6% at roughly 40 minutes per task where GPT-5.6 Sol scored 65.7% at roughly 75 minutes. The Codex harness was updated alongside it, making task completion about 1.9x faster than the Sol experience on Mind2Web.

A practical tip: when briefing Work, describe the deliverable, not the answer. “Analyse this data” performs measurably worse than “read this CSV, plot a monthly breakdown and produce a one-page summary document”. Give it your files, constraints and acceptance criteria up front.

Using Astra in Codex and the API

Codex

Codex is a separate view in the desktop app, selected from the top-left menu. It works with local folders, repositories, terminals and developer tools, and its history stays separate from ChatGPT history. It is not selectable on web or mobile.

The important Codex change shipping with Astra is context management. Previously, when the context window filled, the model compressed its work into a single summary (compaction), and details like why a fix failed got lost. Astra instead keeps notes across context windows; earlier windows stay searchable, so it can retrieve a test result even if that detail never made it into the notes. The feature is experimental for now: enable it in your config.toml, and it becomes the default for Astra in the coming weeks.

One more behavioural change: when a gap could change the outcome, Astra asks asynchronously. It raises the question and keeps working on whatever does not depend on your answer. If you never reply it proceeds with sensible assumptions, but it waits on consequential decisions.

API

The model id is gpt-6-astra. The spec sheet:

A minimal call:

from openai import OpenAI

client = OpenAI()

response = client.responses.create(
    model="gpt-6-astra",
    reasoning={"effort": "medium"},
    input="Explain why this Arduino sketch triggers a watchdog reset, then fix it.",
)

print(response.output_text)

The model is also served through Microsoft Azure and AWS Bedrock. Eligible API customers can run Astra with Zero Data Retention.

Rate limits scale by tier: Tier 1 starts at 500 requests and 500K tokens per minute, Tier 5 reaches 15K requests and 40M tokens per minute.

Keeping the Bill Under Control

A $10 in / $50 out tariff adds up quickly if you are careless. The full rate card:

Three practical rules:

1. Use the cache. Put fixed system prompts and unchanging context at the front of the request. A cache read costs one tenth of fresh input.
2. Mind the 272K threshold. Requests above 272,000 input tokens are billed at 2x input and cache rates and 1.5x output, applied to the whole request. A million-token window is not an invitation to paste the entire repository.
3. Match effort to the job. max is not required for every question; you pay for the extra score in latency and tokens.

Count what a prompt actually costs you with our token counter, and compare monthly spend across models, Astra included, with the LLM cost calculator.

Not Every Job Needs Astra

Being honest: this is an expensive model and it is overkill for most tasks.

• Desktop and browser automation, CAD, scientific terminal work: where Astra is clearly ahead. The time saved matters more than the score gap.
• Everyday coding: the margin over rivals is 3-4 points. If budget matters, cheaper models still make sense.
• High-volume classification, labelling, summarisation: not Astra’s lane. Gemini 3.8 Flash or GPT-5.6 Luna do the same job at a tenth of the price.
• Cybersecurity: the standard-access version refuses vulnerability discovery work; check Daybreak access before picking a model at all.

For the score detail and its footnotes, see the GPT-6 Astra launch post.

Frequently Asked Questions

Q: How do I use GPT-6 Astra? A: It depends on your plan. On Pro, Business and Enterprise it appears in the ChatGPT model picker under the Pro tier, labelled GPT-6 Pro. On Plus it does not appear in regular Chat at all; access comes through ChatGPT Work and Codex. Developers call it in the API with the model id gpt-6-astra.

Q: Is GPT-6 Astra free? A: No. It is not included on Free or Go plans, whose default model is GPT-5.6 Luna. There is no free Astra tier or trial; the cheapest entry point is the limited allowance inside a Plus subscription’s Work and Codex.

Q: I have Plus, why is Astra missing from my model picker? A: Because Plus does not get it in Chat. Per OpenAI’s help centre, Plus plans include Astra in ChatGPT Work and Codex. To find Work, open the desktop app, select ChatGPT from the top-left menu, then switch to Work with the toggle at the top. Work is itself rolling out gradually, so it may not be enabled on your account yet.

Q: Are GPT-6 Pro and GPT-6 Astra the same thing? A: Yes. In the Chat interface OpenAI presents Astra under the GPT-6 Pro label. The help centre wording is “GPT-6 Pro, powered by GPT-6 Astra”.

Q: What are the GPT-6 Pro message limits? A: 200 messages per week on Pro $200, 50 per week on Pro $100, 50 per week on Business Premium and 15 per month on Business Standard. Those apply to Chat; Work and Codex have separate allowances. On Pro $100 and the Business plans the allowance is shared with GPT-5.6 Sol Pro.

Q: What do I need to run Astra in Codex? A: Codex CLI 0.153.0 or newer and an up-to-date ChatGPT desktop app. Signing in with your ChatGPT account draws on your plan’s allowance; using your own API key bills at API rates.

Q: What is GPT-6 Astra’s context window? A: 1,050,000 tokens, with a maximum input of 922,000 and a maximum output of 128,000 tokens. The knowledge cutoff is 30 April 2026. Requests above 272,000 input tokens are billed at 2x input and 1.5x output.

Q: Why is Astra eating my allowance so fast? A: OpenAI states that Astra consumes allowance faster than GPT-5.6 Sol on the same work. Consumption depends on the task, input and output size, the reasoning level you choose and Fast mode. The gap widens on long-context, high-effort requests.

Q: Will buying credits get me Astra sooner? A: No. Credits only cover additional usage after your allowance runs out; they do not grant early access during the rollout.

Stay well… 🙂