THE VERDICT
● Claude Opus 5 for anything you are deciding this week. $5 and $25 per million, shipping since July, independently tested.
● GPT-6 Astra once pricing and independent benchmarks exist. Neither did at launch.
● The honest position: you cannot compare these properly yet, and anyone telling you otherwise is quoting a launch post.
What we know about each
| GPT-6 Astra | Claude Opus 5 |
|---|
| Released | 3 September 2026 | July 2026 |
| API pricing | Not published at launch | $5 in / $25 out per million |
| Independent benchmarks | None yet | Available |
| Availability | Phased. Plus, Pro, Business, Enterprise, API and AWS over coming days | General. Default on Claude Max |
| Capability restrictions | Cyber capability gated behind Daybreak | None of this kind |
THE ASYMMETRY
Opus 5 has two months of people using it on real work and publishing what they found. Astra has a launch post, an executive quote about a generational leap, and a safety disclosure.
The safety disclosure is the most credible thing in the announcement, because it is the only claim made against the company's own interest.
What to do this week
| If you are... | Do this |
|---|
| Choosing a production model now | Opus 5. Known price, known behaviour |
| On ChatGPT Plus or Pro | Astra arrives in coming days at no extra cost. Try it then |
| Budgeting API spend | Wait. There is no Astra number to budget against |
| Reading launch-day comparisons | Check whether any figure came from someone other than OpenAI |
FAQ
Which is better?
Unknown. Astra launched on 3 September with no published pricing and no independent benchmarks. Opus 5 has been in production since July at $5 and $25 per million.
What does the Daybreak gating mean for me?
If you are not in that programme, you get Astra without its advanced cyber capability. For most work that is irrelevant.
When should I revisit this?
When API pricing is published and a third party has benchmarked it. Both are usually within weeks of a launch.