OpenAI’s GPT-6 Astra represents a substantial release, achieving record views and likes within its first nine hours. The model is positioned as a new flagship, designed for computer use, software engineering, and other complex tasks. OpenAI claims Astra achieves state-of-the-art performance, surpassing GPT-5.6 Sol on several benchmarks, including a 1.9x speed increase.
Pricing for Astra is notably higher, at $10 per 1 million input tokens and $50 per 1 million output tokens. However, OpenAI highlights a faster tier with pricing of $20 and $100 per million tokens, offering a potential 2.5x speed improvement. Alongside the model, OpenAI introduced features like Codex’s ability to independently work while asking questions, and an experimental context feature for extended task management.
Independent evaluations paint a nuanced picture. ArtificialAnlys found Astra comparable to Claude Opus 5 and Fable 5, with a 70% more token-efficient performance than GPT-5.6 Sol. Despite these gains, concerns remain regarding monitorability and evaluation. The benchmark data reveals a 2.5x higher token price, making Astra 75% more expensive per task than its predecessor at maximum effort.
Furthermore, the launch was accompanied by initial user feedback. Reports indicate delays, confusion around access, and frustration regarding influencer access before paying users. OpenAI responded with “banked resets” to compensate for lack of access. The release of the deployment safety material sparked significant discussion due to its emphasis on decreased chain-of-thought monitorability.
Source: https://www.latent.space/p/ainews-gpt-6-astra-openais-biggest
