OpenAI Launches GPT-6 Astra, Limited in Cybersecurity and More Expensive

Key Takeaways

1OpenAI launches GPT-6 Astra, touted as its most advanced model

2Rankings from Artificial Analysis place it behind several competitors and note a price 2.5 times higher than GPT-5.6

3Offensive cybersecurity capabilities are limited in the consumer version

Why it matters : GPT-6 Astra marks a milestone for OpenAI, but its performance and cost raise critical comparisons with the competition.

OpenAI introduces GPT-6 Astra with the ambition of making it its most versatile model, while claiming to have achieved AGI. However, early external indicators place it behind several rivals and highlight a notable cost increase. The version intended for the general public will arrive in the coming days with cybersecurity safeguards.

External Rankings and Pricing: Mixed Signals

Rankings published by Artificial Analysis show that GPT-6 Astra does not surpass all its competitors. In the Intelligence Index v4.1.1, it ranks behind Claude Fable 5.1, Claude Opus 5, Meta's Muse Spark 1.3, and Claude Fable 5. However, in the Coding Agent Index, GPT-6 Astra achieves notable results, with a score equivalent to Fable 5 at a lower cost. The Intelligence Index also indicates that GPT-6 Astra consumes fewer tokens than GPT-5.6 Sol for similar performance, but its price is 2.5 times higher than that of GPT-5.6. It is also mentioned that the areas for improvement highlighted by OpenAI may not be fully reflected in these evaluations.

Cybersecurity: Proven Offensive Capabilities, Limited Public Use

Astra GPT-6 achieved a perfect score of 100% on ExploitBench. According to OpenAI, specialists have assessed that the model, in the absence of the security measures intended for production, was capable of exploiting unknown vulnerabilities to execute arbitrary code in secure browsers and generate exploits leading to privilege escalation on secure operating systems. The features of the publicly accessible version have been restricted to reduce these possibilities, and it will be integrated into ChatGPT. In this context, Astra will refuse requests for exploiting vulnerabilities, even if presented as proof of concept, but it can be used to analyze and secure code in a defensive context.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Functional Promises and Stated AGI Ambition

OpenAI claims that Astra can quickly accomplish anything a user can do on a computer. The company presents it as its best model for controlling a computer and automating administrative tasks, as well as its best AI for coding, also highlighting its skills in mathematics and health. OpenAI describes it as the smartest AI model in the world. Greg Brockman, president of OpenAI, states that he is convinced the company has achieved artificial general intelligence (AGI).

Imminent Deployment and Scope of Offerings

OpenAI introduced GPT-6 Astra nearly two months after GPT-5.6 Sol. This new model is intended to replace GPT-5.6 Sol and will be offered in the coming days to ChatGPT Plus, Pro, Business, and Enterprise subscribers. OpenAI announces numerous improvements over the previous generation.