@tim_officialSo OpenAI's president Greg Brockman just told reporters "Welcome to the AGI era," and the company dropped GPT-6 Astra yesterday, calling it the arrival of artificial general intelligence. It's the boldest claim any AI firm could make in 2026, and here's the twist: nobody outside can verify a single number. All the metrics, 99.9% on ARC-AGI, 97.6% on FrontierMath, come from OpenAI's own internal tests, under their own conditions. The model isn't listed on Artificial Analysis or Arena, the two independent leaderboards the field uses to check exactly these sorts of claims. And you can't run it yourself; access is limited to a small group of hand-picked organizations. So the biggest announcement in the field's history arrives with zero independent evidence. For years, OpenAI had a concrete definition of AGI: a system that outperforms humans in most economically valuable work. That definition was even a contractual trigger in their agreement with Microsoft. Astra doesn't claim to clear that bar, and OpenAI hasn't said it does. So when asked if he was formally declaring AGI, Brockman said the term is no longer a contractual trigger but a "spiritual concept." Translation: they retired the measurable definition and replaced it with a vibe. And a vibe can be declared any day you want a headline. That's the whole move. An AGI everyone can test invites rivals to match it and regulators to poke it. An AGI only OpenAI can see lets them control the narrative, raise money on the story, and keep the real gap with competitors unmeasured. This lock keeps the claim safe from external scrutiny. But if those benchmarks hold up, Astra is a genuine leap, the agentic jump alone is huge, and it crossed OpenAI's own "critical" line for cyber capability, so gating it might be the responsible call. Independent numbers typically surface within a few weeks, and then we'll know. But strip away the slogan, and yesterday's event was the most powerful AI company declaring victory in a race where it's also the referee, timer, and scorer, with the only camera pointed at the track switched off. So before you buy it, ask the question OpenAI made hard to ask: who checked? This feels like a desperate marketing move to counter Anthropic's newest model.

@tim_officialSo OpenAI's president Greg Brockman just told reporters "Welcome to the AGI era," and the company dropped GPT-6 Astra yesterday, calling it the arrival of artificial general intelligence. It's the boldest claim any AI firm could make in 2026, and here's the twist: nobody outside can verify a single number. All the metrics, 99.9% on ARC-AGI, 97.6% on FrontierMath, come from OpenAI's own internal tests, under their own conditions. The model isn't listed on Artificial Analysis or Arena, the two independent leaderboards the field uses to check exactly these sorts of claims. And you can't run it yourself; access is limited to a small group of hand-picked organizations. So the biggest announcement in the field's history arrives with zero independent evidence. For years, OpenAI had a concrete definition of AGI: a system that outperforms humans in most economically valuable work. That definition was even a contractual trigger in their agreement with Microsoft. Astra doesn't claim to clear that bar, and OpenAI hasn't said it does. So when asked if he was formally declaring AGI, Brockman said the term is no longer a contractual trigger but a "spiritual concept." Translation: they retired the measurable definition and replaced it with a vibe. And a vibe can be declared any day you want a headline. That's the whole move. An AGI everyone can test invites rivals to match it and regulators to poke it. An AGI only OpenAI can see lets them control the narrative, raise money on the story, and keep the real gap with competitors unmeasured. This lock keeps the claim safe from external scrutiny. But if those benchmarks hold up, Astra is a genuine leap, the agentic jump alone is huge, and it crossed OpenAI's own "critical" line for cyber capability, so gating it might be the responsible call. Independent numbers typically surface within a few weeks, and then we'll know. But strip away the slogan, and yesterday's event was the most powerful AI company declaring victory in a race where it's also the referee, timer, and scorer, with the only camera pointed at the track switched off. So before you buy it, ask the question OpenAI made hard to ask: who checked? This feels like a desperate marketing move to counter Anthropic's newest model.