[AINews] GPT-6 Astra: OpenAI’s biggest LLM launch of all time

OpenAI put GPT-6 Astra out as its new flagship model, and the launch started making noise almost immediately. The first 9 hours were enough to rack up 36 million views and 164,000 likes, which Latent Space says makes it OpenAI’s most successful launch since Sora, and bigger than GPT-4 or GPT-5.

But the model itself was only half the story. OpenAI pitched Astra as its “most intelligent and aligned model yet,” with a focus on computer use, software engineering, math and science, polished office work, and cybersecurity. Rollout was staged: a small set of organizations first, then ChatGPT Plus, Pro, Business, Enterprise, the API, and AWS over the following days. Pricing was also spelled out clearly, at $10 per 1 million input tokens and $50 per 1 million output tokens for the standard model, or $20 and $100 for the fast version.

The launch didn’t exactly glide. Users reported delays, a broken or late blog post, unclear timing for access, and frustration that some influencers got in before paying users. OpenAI responded by offering “banked resets” for days when paid users couldn’t access Astra. At the same time, the company released safety material that drew heavy attention because it described both stronger alignment and lower chain-of-thought monitorability.

That tension ran through the benchmark debate too. OpenAI’s own numbers were eye-popping: 99.9% on ARC-AGI-3, 98% on FrontierMath Tier 4, and 100% on ExploitBench. But outside evaluators were more mixed. Artificial Analysis said Astra looked efficient in coding workflows, yet not dominant across the board, and flagged regressions in some tests. Epoch called it a record, but not a clean break from the trend line. The broad read from the feed was simple: Astra looks strong, but the cost, harness, and safety story matter just as much as the headline scores.

Astra’s best reception came from computer use, long-horizon work, scientific reasoning, and 3D generation. Its sharpest criticism came from people worried about monitorability, benchmark saturation, and whether the alignment story is really a fix or just better packaging.