On this page
The short version
- 01Claude Fable 5.1 (1 September) and GPT-6 Astra (3 September) are the new flagship AI models from Anthropic and OpenAI.
- 02On paper they cost the same — US$10 per million input tokens and US$50 per million output — but real bills depend on caching, settings and the tools around them.
- 03For most small businesses the better question isn’t which model wins a benchmark, but which one fits the software you already use.
In the first week of September 2026, Anthropic and OpenAI released new flagship AI models within two days of each other: Claude Fable 5.1 on 1 September and GPT-6 Astra on 3 September. The comparisons started immediately — and they haven’t agreed with each other since.
If you run a business rather than an AI lab, here’s what actually matters.
The short comparison#
| Claude Fable 5.1 | GPT-6 Astra | |
|---|---|---|
| Maker | Anthropic | OpenAI |
| Released | 1 September 2026 | 3 September 2026 |
| Where you can use it | Claude apps, Claude API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Azure | ChatGPT paid plans, OpenAI API, Amazon Bedrock, Microsoft Azure |
| List price per million tokens | US$10 input · US$50 output | US$10 input · US$50 output |
| Cached input per million tokens | US$0.25 | US$1.00 |
| Context window | About 1 million tokens | About 1.05 million tokens |
A token is a chunk of text — roughly three-quarters of a word. A million tokens is a very long document, or a great deal of back-and-forth.
Where each one stands out#
Both companies claim the lead, and each publishes benchmarks that favour its own model. Independent testers put them close, with the winner changing depending on the task and the settings used.
Claude Fable 5.1
Anthropic positions Fable 5.1 as its most advanced model for coding and knowledge work, built for long-running, multi-step tasks. The practical headline is price: cached input fell to US$0.25 per million tokens, which matters for repetitive, agent-style work where the same instructions and documents are reused. Anthropic also released Claude Mythos 5.1 — the same model with different safeguards — only to vetted security and life sciences programmes.
GPT-6 Astra
OpenAI describes Astra as its most capable model for complex end-to-end work, pushing hardest on computer use — operating software and browsers on your behalf — alongside coding and scientific reasoning. It launched in stages, starting with organisations in OpenAI’s cyber security programme before reaching paid ChatGPT plans and the API.
Astra has also drawn criticism. It uses a reasoning technique that makes the model’s step-by-step ‘thinking’ harder for researchers to inspect — a trade-off OpenAI has acknowledged.
What this means for a small business#
Most businesses won’t call these models directly. They’ll meet them inside tools they already pay for — ChatGPT or Claude subscriptions, Microsoft and Google products, and specialist apps built on top. So the practical questions are simpler than the benchmark tables:
- 01What do you already use? If your team lives in one ecosystem, start with the model that’s available there.
- 02What’s the job? Drafting, summarising and research work well on either. Automating tasks across your systems is where differences show.
- 03Where does your data go? Check the plan’s data terms before anyone pastes client information in — business and consumer plans often differ.
- 04Who’s accountable? If AI output reaches customers, someone should review it, and your privacy obligations still apply.
Our take#
Both are genuinely capable, and the gap between them is smaller than the launch-week noise suggests. For most of our clients the win isn’t switching models — it’s choosing one well-governed tool, connecting it properly to the systems you already run, and measuring whether it saves time. That’s the work we do in AI automation.
The best model for your business is usually the one your team will actually use safely — not the one that topped a chart this week.
Common questions
Which is better, Claude Fable 5.1 or GPT-6 Astra?
It depends on the task. Each company’s own benchmarks favour its model, and independent testing puts them close. Fable 5.1 is often preferred for long, multi-step knowledge and coding work; Astra is pitched strongly at computer use, maths and science.
How much do Claude Fable 5.1 and GPT-6 Astra cost?
Both list at US$10 per million input tokens and US$50 per million output tokens through their APIs. Cached input differs — US$0.25 for Fable 5.1 and US$1.00 for Astra — so real costs vary with how you use them. Subscription plans are priced separately.
Can Australian businesses use these models through Microsoft Azure or AWS?
Yes. Both are offered through Microsoft Azure and Amazon Bedrock, alongside their makers’ own apps and APIs. Specific features and regions can vary, so check before you build.
Sources
- 01Anthropic — Introducing Claude Fable 5.1 and Claude Mythos 5.1
- 02TechCrunch — OpenAI launches Astra, its powerful (and controversial) new model
- 03DataCamp — GPT-6 Astra vs Claude Fable 5.1: Benchmarks and Pricing
- 04MindStudio — GPT-6 Astra vs Fable 5.1: What They Actually Cost to Run
- 05Artificial Analysis — GPT-6 Astra vs Claude Fable 5.1
