A proposal can look finished while still hiding an unsupported promise, a wrong total or a missing deliverable. For a solo operator, checking those details is often the real work.
GPT-6 Astra makes that checking stage a useful place to start. OpenAI positions the model for demanding work across reasoning, research, software and documents. The practical question is whether it can help you move a real assignment from brief to an output you can confidently use.
Information checked: 5 September 2026. This is a source-based practical guide, not a hands-on performance review. The workflow and numerical examples below are illustrative.
What GPT-6 Astra means for ChatGPT users
The official model name is GPT-6 Astra. OpenAI’s weekly update describes its use in Codex and ChatGPT Work for complex tasks involving code, applications and research, including producing documents, spreadsheets and presentations. Access is account-dependent: select Astra when it becomes available in your model selector. Enterprise access also depends on rollout eligibility and administrator enablement. Do not assume that seeing an announcement means your account already has access. OpenAI’s product update
The model and the application around it have different responsibilities. A model can reason about a spreadsheet; accessing your spreadsheet requires the relevant files, tools and permissions. Start with material you can safely supply and check which capabilities your actual environment exposes.
Give Astra a finished-work assignment
OpenAI’s model guidance describes support for multistep work and mid-turn steering, which allows additional instructions during an ongoing task in supported implementations. It also notes that Astra may ask more clarifying questions and that instructions in skills or project files can influence its behaviour. Those are reasons to make your assignment precise. GPT-6 Astra model guidance
A useful first assignment is a proposal readiness check. It has a defined business purpose, evidence you can inspect and mistakes that can be caught before anything reaches a client.
Imagine an independent consultant preparing a fixed-scope service proposal. The inputs are an anonymised client brief, an approved service description, a rate sheet and a proposal template. The desired output is a completed draft plus a short record of unresolved issues.
That package can become a reusable business asset. Each future proposal begins with established scope rules and a known review process, reducing the work of rebuilding the same structure.
A five-step proposal workflow
1. Prepare the evidence. Gather the current versions of the four inputs. Remove personal details and confidential commercial information unless your organisation permits their use in the chosen environment. Label the rate sheet’s currency and effective date. Keep originals intact and work on copies.
Missing inputs are not a licence to invent them. If the rate sheet does not state tax treatment, the draft should flag that gap.
2. Define what acceptance means. Write down the tests before generating the proposal: every service must come from the approved description; every price must trace to the rate sheet; exclusions must be visible; and unsupported delivery promises must be marked for a decision.
Use a request such as:
> Prepare a proposal draft from these four files. Preserve the template’s headings. Use only the supplied services and prices. Identify conflicting or missing information. Check the arithmetic and list the inputs behind each total. Make routine formatting decisions yourself. Ask me about changes to price, scope or commitments. Return the draft and an issues list. Do not send it or change the source files.
This is an illustrative prompt, not a tested guarantee. Its value is that it makes the expected output and decision boundaries visible.
3. Inspect the first result. Open the actual document. Check its tables, totals, line breaks and missing fields. Follow its references back to the inputs. A statement that checks passed is not a substitute for seeing what was checked.
For arithmetic, independently reproduce a sample calculation using a calculator or spreadsheet. For scope, compare each deliverable with the approved service description.
4. Make one realistic revision. Change a requirement: remove a service, shorten the engagement or alter a quantity. Ask Astra to update the proposal and identify every affected section.
Then inspect for stale references. Did the total change while the payment schedule stayed the same? Did a removed service survive in the executive summary? This reveals whether the workflow holds together through revision.
5. Save the reusable parts. Retain the clean template, source checklist, assignment prompt and acceptance criteria. Keep a brief error log. Assign someone to maintain prices and service descriptions; a polished workflow built on old inputs still produces old answers.
Measure the cost of an accepted proposal
For API users, OpenAI lists standard Astra text pricing at US$10 per million input tokens and US$50 per million output tokens. Requests above 272,000 input tokens have higher rates. These are API prices, not ChatGPT subscription prices. Official model specifications and pricing
At those standard rates, an illustrative request using 20,000 uncached input tokens and 5,000 billable output tokens costs US$0.45 for those tokens. This excludes tool charges, extra calls and other applicable costs; billable output can include reasoning, so visible text alone is not a complete cost measure.
For most individuals, review time deserves equal attention. If a manual proposal takes 90 minutes and an AI-assisted attempt takes 15 minutes of preparation, 20 minutes of checking and 10 minutes of corrections, the illustrative saving is 45 minutes. It is not a measured Astra result.
Track preparation, waiting, review, corrections and fees across several comparable proposals. Compare with your existing method or a cheaper model using the same evidence and acceptance criteria. Keep Astra for assignments where its contribution justifies the total cost.
Where responsibility stays with you
Astra cannot decide whether a commercial promise is sensible merely because it can write it clearly. You own pricing, client commitments and final release.
Review your plan’s data-handling and retention settings before uploading client material. Limit connected access, keep backups and avoid granting permission to send messages merely to prepare a draft. Recheck templates and integrations when tools change.
The immediate opportunity is a repeatable proposal process that produces fewer unresolved details and more usable work. That may free time for delivery or sales; it does not guarantee additional income.
Choose one recent, anonymised brief. Rebuild its proposal, introduce a revision and inspect the result against your original. Keep the workflow only if the accepted output earns back the effort of supervising it.
Sources and disclosure
- OpenAI: What’s new in ChatGPT and Codex
- OpenAI: GPT-6 Astra model guidance
- OpenAI: GPT-6 Astra specifications and API pricing
AI assisted with research, drafting and the original editorial illustration. Examples and savings calculations are illustrative; no independent Astra benchmark or client trial is claimed. Product access, pricing and capabilities may change.
