GPT-6 Astra in ChatGPT Work & Codex: Which OpenAI Plan Your Agency Should Buy
Status as of September 4, 2026: GPT-6 Astra is live. OpenAI's developer account confirmed Astra is available in the OpenAI API (model ID gpt-6-astra) and in ChatGPT Work and Codex for all Pro, Enterprise, and Business Premium users. Pro, Enterprise, and Business Premium plans also include the GPT-6 Astra Pro tier. Enterprise administrators can enable Astra for their workspace — access is off by default at launch. Plus and Business chat are the remaining rollout surfaces. This page is the buying decision for agencies: which plan, what Astra Pro costs in Work/Codex, and what agent access actually changes in delivery.
What happened. GPT-6 Astra launched September 3, 2026 as OpenAI's computer-use flagship and its first Critical-tier cybersecurity model, starting with Trusted Access and Daybreak organizations. On September 4, OpenAI extended it to the API and to ChatGPT Work and Codex for Pro, Enterprise, and Business Premium (OpenAI developer account; HuggingNews), and OpenAI exec Tibor Sottiaux said Pro and Business subscriptions were getting Astra first in Work/Codex with Plus to follow. 9to5Mac's September 4 update put it in plan terms: Astra is "rolling out to Business and Pro customers on the $100/month or $200/month plan" — before the $20/month Plus plan — and quoted OpenAI: "Astra usage is included within the existing subscription allowances — users and businesses will also be able to purchase credits for additional usage. Users on the Pro, Business, and Enterprise plans will also get access to GPT-6 Astra Pro. Enterprise administrators can enable Astra for their workspace; access is off by default at launch." This is the fast-follow to our GPT-6 Astra model comparison and computer-use guide, which has the full model-vs-model table.
The access reality: it is live, but account-by-account rollout still lags
Before the plan math, one operational warning. OpenAI's own model page still described customer access as rolling out "over the next several days" at the September 4 cutoff, and Kingy's access matrix separates a documented release from access that has actually reached an account. The practical checklist for an agency:
- In ChatGPT/Work: open the model picker in the specific workspace you intend to use. OpenAI displays Astra in ChatGPT as "GPT-6 Pro" — the name "Astra" may not appear in the picker (Kingy Brief).
- In Codex: update Codex CLI to 0.153.0 or newer, then check the model picker; Astra is rolling out separately across Work and Codex.
- In the API: confirm
gpt-6-astraappears in the model list available to your project before moving production traffic. - Credits do not buy rollout priority. OpenAI says credits extend usage only after Astra reaches your account.
Agency access at a glance (September 4, 2026 cutoff)
| Plan | List price | GPT-6 Pro in Chat (Astra) | Astra in ChatGPT Work/Codex | Astra Pro tier |
|---|---|---|---|---|
| Plus | $20/month | Not included at Sept 4 cutoff (OpenAI plan guide); OpenAI says Plus follows the $100/$200 plans | Limited Astra usage when rollout reaches the account; optional credits after included amount | No |
| Pro $100 | $100/month | Included — ~50 Astra messages/week shared with GPT-5.6 Sol Pro (Kingy Brief) | Full existing Work/Codex allowance for Astra | Yes (Pro plan) |
| Pro $200 | $200/month | Included — ~200 Astra messages/week; GPT-5.6 Sol Pro separately ~170/day, both capped at ~200/day | Full existing Work/Codex allowance for Astra — the largest single-seat allowance | Yes (Pro plan) |
| Business Standard | Business seat pricing (below Premium) | Included on rollout — ~15 Astra messages/month shared with Sol Pro | Limited Astra usage; optional credits after included amount | No |
| Business Premium | $100/user/month (list; see price note below) | Included — ~50 Astra messages/week shared with Sol Pro | Full existing Work/Codex allowance for Astra, per user, with workspace admin controls | Yes (Business plan) |
| Enterprise | Custom (rate card) | Included — limits through Enterprise controls and the ChatGPT rate card | Full allowance per Enterprise controls — but admins must enable Astra per workspace | Yes (Enterprise plan) |
Source: Kingy Brief access matrix (updated Sept 4, 2026, 12:45 p.m. PT) for message allowances and Work/Codex entitlement; OpenAI statement as quoted by 9to5Mac for Astra Pro on Pro/Business/Enterprise and admin-enablement. Chat allowances are separate meters from Work/Codex credit allowances.
Price note — Business Premium at $100/user/month
The earlier $100 anchor for Business Premium was flagged as uncertain, so here is the verification trail as of September 4, 2026: 9to5Mac's rollout update reports Astra reaching "Business and Pro customers on the $100/month or $200/month plan" — Business on the $100/month plan and Pro on the $200/month plan. That corroborates the $100/user/month Business Premium figure, and it matches OpenAI now also listing a Pro $100 tier in its Astra plan guide (Kingy Brief). Caveat: neither source reproduces OpenAI's pricing-page line item for Business Premium, and workspace plans are frequently billed annually or discounted by seat count — confirm the current list price on OpenAI's pricing page before you budget a team rollout.
The decision: Pro $200/month vs Business Premium $100/user/month
Now that Astra runs in Work/Codex on real subscriptions, the agency question is structural: do you buy individual Pro seats, or per-user Business Premium seats in a shared workspace?
- Solo owner-operator doing the delivery → Pro $200/month. You get the largest documented single-seat allowance: ~200 Astra chat messages per week plus your plan's full Work/Codex allowance for agent runs. If your Astra use is mostly Work/Codex and light on chat, the Pro $100 tier gets the same full Work/Codex Astra entitlement at half the price (Kingy Brief) — the difference is the chat allowance (~50 vs ~200 messages/week).
- Multi-person agency where every builder needs Astra → Business Premium at $100/user/month. The same full Work/Codex Astra entitlement per user, under workspace admin control (who can enable which model), with per-seat chat allowances (~50 Astra messages/week shared with Sol Pro). Two Pro seats ($400) vs two Business Premium seats ($200) is the kind of gap that decides the stack once headcount passes two or three. Business Premium is the team default if your builders can live inside Work/Codex allowances rather than the biggest Pro chat meter.
- Compliance-driven delivery or large seat counts → Enterprise. Astra is admin-enabled and off by default — an Enterprise admin must turn it on per workspace, then set model permissions and limits through Enterprise controls and the rate card. That is a feature for agencies selling into regulated clients, and a trap if you assume Astra is on because the contract says Enterprise.
- Plus or Business Standard → do not buy for Astra. Plus gets only limited Work/Codex Astra usage (and no GPT-6 Pro chat at the Sept 4 cutoff); Business Standard's chat allowance is ~15 messages/month. If Astra agent work is the deliverable, these are evaluation tiers, not production seats.
Astra Pro and the credit economics in Work/Codex
OpenAI's own framing: Astra usage is included within existing subscription allowances, and users and businesses can purchase credits for additional usage. Work and Codex meter Astra separately from chat messages, at published credit rates (OpenAI rate card via Kingy Brief, Sept 4):
- 250 credits per 1M input tokens · 25 credits per 1M cached input tokens · 1,250 credits per 1M output tokens. At API parity (25 credits = $1), that is $10 / $1 / $50 per 1M — the same headline prices as the API, denominated in credits.
- Fast mode costs 2.5x the Standard credit rate in Work and Codex — separate from the API, where Fast is 2x Standard. Example from OpenAI's rate card: a request with 100K uncached input + 20K output consumes roughly 50 Standard credits (~$2) or 125 Fast credits (~$5) in Work/Codex; the same token mix is ~$2 API Standard or ~$4 API Fast.
- The 272K-token cliff applies. Prompts above 272K input tokens reprice the entire request at 2x input/cache and 1.5x output rates (the $20 / $2 / $75 per 1M structure documented on the model page) — long desktop sessions that load whole documents are where this bites. See the model comparison for the full pricing structure.
- Credits are overage, not early access. They extend usage only after rollout reaches your account. Do not buy credits expecting them to unlock Astra faster.
- Chat messages are a separate meter. The ~50/week (Pro $100 and Business Premium) and ~200/week (Pro $200) figures are Chat allowances shared with GPT-5.6 Sol Pro, not Work/Codex credit allowances. An agency running heavy Codex agent work should size on the credit meter, not the chat number.
What "Astra Pro" means here. OpenAI's September 4 statement puts the Astra Pro tier on Pro, Business, and Enterprise plans — included with those plans, not a separate consumer SKU. What OpenAI had not published at this cutoff is a separate credit rate sheet distinguishing Astra vs Astra Pro inside Work/Codex: the published 250/25/1,250 rates are the Astra Work/Codex rates. If your workspace rate card lists an Astra Pro line, read the workspace rate card rather than assuming the standard Astra credit meter applies. Either way, the discipline from our AI agent workload routing guide holds: price per task, not per token — Astra's argument is fewer tokens and fewer retries per task.
Enterprise: admin-enabled, off by default
Enterprise access to Astra is real and it is not automatic. OpenAI's statement is explicit: "Enterprise administrators can enable Astra for their workspace; access is off by default at launch." For an agency on an Enterprise workspace (your own, or a client's), that means:
- An admin must enable the model per workspace and set which users can reach GPT-6 Pro / Astra, then confirm it appears in the model picker. Kingy's matrix: "Access depends on workspace model permissions; current limits are handled through Enterprise controls and the ChatGPT rate card."
- Rollout is phased even inside Enterprise — "limited set of organizations" language from OpenAI's release notes still applied at the September 4 cutoff, and account-level arrival can lag the announcement.
- Enterprise also carries the Daybreak question (below): Daybreak access is provisioned separately, per project, with its own administrator enablement.
What Work/Codex agent access changes for agency delivery
Work and Codex are where agencies already run agent builds and client automations, so Astra there is not just another model — it changes the delivery surface:
- Computer use inside the harness you bill from. Astra upgrades ChatGPT and Codex desktop with background computer use — operating a Mac or other desktop while staff keep working — with OpenAI reporting the harness is nearly 2x faster at computer use with Astra, and that the same optimizations speed GPT-5.6 Sol tasks by ~60% (OpenAI; Ari Weinstein via 9to5Mac). Background computer use turns admin operations (scheduling, form intake, listing management, research, data entry) into agent deliverables under your normal delivery tooling rather than a custom API stack.
- Codex context gets a memory upgrade. In Codex, Astra can keep notes across context windows and keep earlier windows searchable, instead of compaction-only summaries — experimental now, and OpenAI says it will become the default for Astra. For agencies running long client refactors or multi-day automations, that directly changes how much re-prompting your operators do.
- Included allowance changes your COGS. A Pro or Business Premium seat covers Astra agent runs inside the plan's Work/Codex allowance — for a small agency that can be cheaper than per-token API billing for the same agent work. Overage is metered at the credit rates above. Re-quote admin-automation retainers on cost per task and include the credit overage line in the SOW.
- Monitoring is part of the product. OpenAI's misalignment monitor reviews agent reasoning, actions, inputs, and outputs across ChatGPT, Codex, and Responses API use, and can pause or end a flagged conversation (stopped API responses may not be resumable). A monitored long-running agent needs a failure path that does not silently retry forever. This is the operational reason our four-checkpoint autonomy framework (approval, expertise, variance, interest) and a desktop-control scope line belong in every computer-use SOW.
- Model-access is now a workspace setting. Because Astra is admin-gated on Business/Enterprise workspaces, delivery teams should put "Astra enabled per workspace + model permissions" on the onboarding checklist — the same discipline as our ChatGPT Work agent rollout coverage.
Where the Critical-tier cyber gate still applies
Astra being live in Work/Codex does not mean the benchmark cyber model is on your subscription. Two separate facts, both from OpenAI:
- Standard production Astra refuses advanced cyber work. The Astra you get in ChatGPT, Work, Codex, and the API is the safeguarded configuration — OpenAI's headline cyber results came from Daybreak Blue, a separate restricted configuration for vetted defenders (proof-of-concept exploit completion 2.4% → 92% with Blue; cyber red-teaming 7.4% → 76.9%). OpenAI says reduced refusals are not available on Astra for most Daybreak customers, and its public
gpt-daybreak-blue-latestalias still maps to GPT-5.6 Sol, not Astra (Kingy Brief). - A "Daybreak" badge is not Astra Blue. Daybreak membership, a Daybreak alias, and Astra-with-Daybreak-Blue are three different things. If your agency sells AI security work, the client-facing question is which configuration the workspace actually holds — see our Daybreak Blue explainer and the Critical-rating history for the full gating story, plus the security vetting checklist if you are picking vendors on Astra claims.
Bottom line for the cyber angle: the gated Critical-tier capability did not become broadly available on September 4 — only the safeguarded production model did. That separation is exactly what our earlier coverage (GPT-5.6 Cyber / Daybreak Red) predicted would persist.
Frequently asked questions
Which ChatGPT plan gives an agency GPT-6 Astra?
As of September 4, 2026, GPT-6 Astra is live in the OpenAI API and in ChatGPT Work and Codex for all Pro ($100 and $200 tiers), Enterprise, and Business Premium users. Pro, Enterprise, and Business Premium plans also include the GPT-6 Astra Pro tier. Plus and Business Standard get only limited Astra usage in Work and Codex, and OpenAI's plan guide did not include GPT-6 Pro (the ChatGPT name for Astra) in Chat for Plus at that cutoff.
Should an agency buy ChatGPT Pro at $200/month or Business Premium?
For a solo owner-operator running heavy Astra agent work, Pro at $200/month gives the largest single-seat allowance (documented at roughly 200 Astra chat messages per week plus the plan's full Work and Codex allowance). For a multi-person agency where every builder needs Astra in a shared workspace with admin control, Business Premium — reported at $100 per user per month in the September 4 rollout coverage — is the per-seat value play. Confirm the current Business Premium list price on OpenAI's pricing page before budgeting. Enterprise is the choice when compliance, SSO, and workspace controls matter; Enterprise admins must enable Astra because access is off by default at launch.
How do Astra Pro credits work in ChatGPT Work and Codex?
OpenAI says Astra usage is included within existing subscription allowances, and users and businesses can purchase credits for additional usage. OpenAI's published Work and Codex credit rates for Astra are 250 credits per million input tokens, 25 per million cached input tokens, and 1,250 per million output tokens (25 credits equal $1 at API parity), with Fast mode at 2.5 times the Standard credit rate in Work and Codex versus 2 times in the API. A 100K-input / 20K-output request consumes about 50 Standard credits or 125 Fast credits. Credits extend usage after rollout reaches your account; they do not provide early access.
When does Plus or Business chat get GPT-6 Astra?
As of September 4, 2026, Plus and Business chat had not been reached by the rollout. OpenAI executive Tibor Sottiaux said Pro and Business subscriptions were getting Astra first in ChatGPT Work and Codex with Plus to follow, and OpenAI's rollout pattern historically lands new models on $100/$200 plans before the $20 Plus plan. OpenAI's detailed plan guide did not include GPT-6 Pro in Chat for Plus at the September 4 cutoff.
Is Astra in ChatGPT Work and Codex the same as Daybreak Blue Astra?
No. The Astra available through standard ChatGPT, Work, and Codex access is the safeguarded production configuration, which refuses advanced cybersecurity work. OpenAI's headline cyber results came from Daybreak Blue, a separate restricted configuration for vetted defenders that remains application- and verification-gated — and OpenAI's current alias for Daybreak Blue (gpt-daybreak-blue-latest) maps to GPT-5.6 Sol, not Astra. If an agency or client needs Astra for security engagements, confirm which configuration the workspace actually has.
An agency that routes models for cost — and scopes computer use for trust
Browse Vetted AI Agencies →Or run Astra plan and credit scenarios through the AI agency pricing calculator first.
Sources
- OpenAI statement (Astra usage included in subscription allowances; credits for additional usage; Astra Pro tier for Pro/Business/Enterprise; Enterprise admin-enable, off by default) — as quoted in 9to5Mac, "OpenAI releasing major upgrade to ChatGPT and Codex with GPT-6 Astra, details here" (Zac Hall, Sept 3, updated Sept 4, 2026)
- OpenAI developer account (@openaidevs) — GPT-6 Astra available in the API and in ChatGPT Work and Codex for all Pro, Enterprise, and Business Premium users (Sept 4, 2026); summarized in HuggingNews, "OpenAI Extends GPT-6 Astra to API, ChatGPT Work and Codex" (Sept 4, 2026)
- OpenAI exec Tibor Sottiaux (@thsottiaux) — Pro and Business subscriptions get Astra first in ChatGPT Work and Codex, Plus to follow (Sept 4, 2026); summarized in HuggingNews
- Kingy Brief, "GPT-6 Astra Access: Plans, API, Pricing, Daybreak" — access matrix updated September 4, 2026, 12:45 p.m. PT (Work/Codex credit rates 250/25/1,250, Fast 2.5x, message allowances, GPT-6 Pro naming in ChatGPT, Daybreak Blue separation)
- OpenAI API docs — GPT-6 Astra model page (pricing, context, access): developers.openai.com
- OpenAI announcement — X post, Sept 3, 2026: x.com/OpenAI
Accuracy note: Availability, allowance, and credit-rate figures verified September 4, 2026 against OpenAI's developer-account statement and the Sept 4 rollout coverage (9to5Mac, HuggingNews, Kingy Brief). The $100/user/month Business Premium price is corroborated by 9to5Mac's Sept 4 report that Astra is rolling out to Business customers "on the $100/month plan" but is not reproduced from OpenAI's own pricing page here — confirm the current list price before budgeting. Account-level rollout can lag OpenAI's announcements; confirm gpt-6-astra appears in your project/workspace model picker before planning production work. Plan names and prices reflect OpenAI's September 2026 plan structure as documented in the cited sources (Plus $20, Pro $100/$200, Business Standard/Premium, Enterprise). Re-verify before building client proposals on any of these numbers.