[AINews] GPT-6 Astra: OpenAI’s biggest LLM launch of all time
Astra’s launch reads as a capability jump with a serious monitorability tradeoff.
Latent Space says OpenAI’s new flagship model posted strong results in computer use, coding, ARC-style tasks, math, and long-horizon work, though outside evals were mixed on whether it leads everywhere. The piece stresses that Astra is pricier per token but often cheaper per completed task because it uses fewer tokens or actions. The most substantial concern in the writeup is that OpenAI’s safety materials and outside commentary point to less legible chain-of-thought monitoring as the model gets stronger. The rollout was also messy, with staged access, delayed comms, and paid users waiting while some early testers already had it. Latent Space's note
Latent Space says OpenAI’s new flagship model posted strong results in computer use, coding, ARC-style tasks, math, and long-horizon work, though outside evals were mixed on whether it leads everywhere. The piece stresses that Astra is pricier per token but often cheaper per completed task because it uses fewer tokens or actions. The most substantial concern in the writeup is that OpenAI’s safety materials and outside commentary point to less legible chain-of-thought monitoring as the model gets stronger. The rollout was also messy, with staged access, delayed comms, and paid users waiting while some early testers already had it. Latent Space's note
score 10