OpenAI has opened the GPT-6 era, and it did not do it modestly. On 3 September it released GPT-6 Astra, the first model of the GPT-6 generation, and the company’s president marked the moment with the line “Welcome to the AGI era.” That is a very large claim to hang on a launch, so let us separate what actually shipped from the framing around it, because the model looks genuinely strong and the AGI talk is doing a lot of work.
What GPT-6 Astra actually is
Astra is pitched as a major model for computer use, browsing, software engineering, science and professional work, in other words the agentic, do-things-for-you category rather than the write-me-an-email one. The headline specs: a context window expanded to 1.05 million tokens, a knowledge cutoff moved forward to 30 April 2026, and a reorganised product line into two tiers, Astra and the higher-end Astra Pro.
| GPT-6 Astra | Detail |
|---|---|
| Context window | ~1.05 million tokens |
| Knowledge cutoff | 30 April 2026 |
| Tiers | Astra and Astra Pro |
| Price | $10 / million input, $50 / million output |
| Aimed at | Computer use, browsing, coding, science, pro work |
The price tells a story
Astra lands at $10 per million input tokens and $50 per million output, which is the exact same pricing as Anthropic’s Claude Fable 5.1, released two days earlier. The matching price is deliberate: the two leaders are anchoring the top of the market at an identical, premium number. For a flagship aimed at heavy agentic work that burns tokens fast, that price is the first thing your finance team will notice, and it makes the cheaper models look even more attractive for everyday jobs.
The friendly sceptic’s corner
Now the AGI claim, said with respect. “Welcome to the AGI era” is a marketing sentence, not a technical one. AGI, artificial general intelligence that matches a human across the board, has no agreed definition and no agreed test, which is precisely why it is such a convenient banner to unfurl at a launch. GPT-6 Astra may well be a genuine step up on computer use and coding. That is a different, more useful claim than “AGI is here”, and conflating the two is how hype outruns reality. Judge it on what it does on your work, not on the adjective.
What it means for you
If you do heavy agentic work, coding, browsing tasks, multi-step research, Astra is worth testing against Claude Fable 5.1, since they now sit at the same price and the choice comes down to which one performs better on your specific tasks. For everyday use, a top-tier flagship at $50 per million output tokens is overkill, and a cheaper model will get most people to the same place. The pattern of 2026 holds: the best model and the right model are rarely the same, and the gap between them is your budget.
What this means
GPT-6 is a real generational step and a real marketing event at the same time, and it takes a little effort to keep those apart. The specs are impressive, the price is steep, and the AGI framing is a claim to watch rather than accept. Test it on your actual workflow, compare it honestly to Fable 5.1 at the same price, and treat “the AGI era” as the slogan it is until the work proves otherwise. (Prices as of early September 2026.)
Related on Top Tool Stack: GPT-6 Astra vs Claude Fable 5.1 · ChatGPT vs Claude vs Gemini
Did you know: GPT-6 Astra and Claude Fable 5.1 launched within two days of each other at the identical price of $10/$50 per million tokens. When the two leaders price a flagship the same, the competition has moved from cost to capability.