OpenAI has unveiled GPT-6 Astra – a new model positioned as closest to general AI

Announcement of GPT-6 Astra

According to НВ — Техно: On September 4, OpenAI officially presented its new flagship development – the GPT-6 Astra model. According to the developers, it significantly outperforms the previous version, GPT-5.6 Sol, in solving complex tasks. OpenAI claims this system is the most intelligent among its models and emphasizes that its capabilities approach the level of artificial general intelligence (AGI), which means AI that can handle a wide range of tasks at the level of a human or even more effectively. For Russian-speaking specialists, this is an important signal: the development is not only aimed at increasing power but also at consciously limiting the application of such systems in risky scenarios.

Test results

During testing, GPT-6 Astra outperformed its predecessor across a number of indicators. Here are the key figures:

• ARC-AGI-3: 98.6%
• Terminal Bench 4.0: 57.7%
• Agent's Last Exam: 59.3%
• ExploitBench: Astra – 100%, GPT-5.6 Sol – 78.5%
• SRE-Bench: Astra shows 88% on the first attempt and 99.2% after four attempts; for GPT-5.6 Sol, these figures are 55.9% and 68.7%, respectively

The company also added that additional cybersecurity limitations have been set for the Astra model. In particular, the system is prohibited from performing dangerous cyber tasks and has safeguards against jailbreak requests.

Currently, access to GPT-6 Astra is open only to a limited number of organizations. In the coming days, the model is expected to become available to ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI and AWS API.

The API access to Astra is priced at $10 per million input tokens and $50 per million output tokens. Greg Brockman, a representative of OpenAI, noted that businesses should consider not just the price for tokens but the actual cost of executing a specific task when choosing a model.

"The launch of GPT-6 Astra demonstrates how quickly artificial intelligence is advancing and how intensely technology companies are competing for new achievements in this field." Greg Brockman, OpenAI

Thanks to enhanced capabilities and high results, the model is capable of impacting many areas – from business and education to scientific research. At the same time, it is important to remember the safety and ethics of AI usage, especially as dependence on such technologies continues to grow.

While OpenAI presented a new model, similar achievements are being demonstrated by SpaceXAI with their Grok 4.6. This system has also shown results comparable to GPT-5.6, underscoring the growing competition in artificial intelligence. It will be interesting to see how new developments will affect the market and what opportunities they will open for users.