A few thoughts on the GPT-6-Astra release.

First of all, all my thoughts and assumptions are based on what OpenAI has presented. I have not tested the model myself yet, so I cannot form my own judgment. Still, the information available points to the following conclusion:

GPT-6-Astra is the Fable k1ller we hoped for and expected. The benchmarks suggest that OpenAI has clearly established its dominance today with Astra, its newly trained base model. That much is clear.

GPT-6 has made extraordinary leaps across key benchmarks. It represents a major leap in AI development overall and shows that there is still no end in sight. What’s more, Sam Altman and Greg Brockman have repeatedly described it as probably AGI. I want to pause on that for a moment: there is no universally agreed definition of AGI. The frontier labs largely differ on which characteristics define it. But I think AGI is a pretty good description of GPT-6. Why? Because we can take the term “general intelligence” seriously. GPT-6 is a general intelligence across numerous areas of work: browser use, computer use, Excel spreadsheets, coding, drug discovery, mathematical research, and so much more. Astra performs at an expert level in all these areas, making it a truly general intelligence.

We will soon see how well it performs in practice. The timing of the release, however, is well chosen and no coincidence either. Just days after the release of the frontier model Fable 5.1, and shortly before Anthropic’s planned IPO, OpenAI is dethroning its fiercest competitor. A humiliation planned well in advance, surely also intended to make Anthropic’s IPO less appealing to potential investors and OpenAI’s subsequent IPO more attractive.

We remember that image from India, when Sam Altman and Dario Amodei did not want to hold hands. Today, OpenAI has taken its revenge. Dario Amodei once left OpenAI and built a competing company. In doing so, he publicly humiliated Sam Altman: Anthropic was not only building stronger models but also generating more revenue. Today, however, OpenAI is celebrating its long-awaited revenge and dethroning its fiercest rival.

We know that GPT-6 will not be OpenAI’s last model of 2026. “Bel” is presumably still being trained and will arrive at the end of the year. 2026 is the turning point that will change everything. Astra shows that AI will enter every imaginable field and workplace, initially assisting with tasks and presumably also handling them independently. Meanwhile, development continues, with no end in sight.

The economy will be transformed, and we should all read the AI2027 blog again to think through all the possible scenarios. Because one thing is clear: Astra is only the beginning. You could say that today marks the birth of real AI.