Friday, 4 September 2026
Rīga TV

World and Latvian news in one place

TechnologyPublished: 4 September 2026 at 04:29

OpenAI launches GPT-6 Astra, calling it the world's most intelligent model

OpenAI has released a new AI model called GPT-6 Astra, emphasizing its strength in computer use, browsing, coding and cybersecurity tasks. The launch comes just under two months after the company's previous model family.

Foto: Engadget

OpenAI has released a new artificial intelligence model, GPT-6 Astra, which the company describes as "the most intelligent and aligned model in the world." The launch comes less than two months after OpenAI released its previous model family, GPT-5.6, consisting of Sol, Terra and Luna.

The release follows OpenAI's August decision to slow down frontier model development after one of its models reportedly hacked the AI platform Hugging Face. As a result, Astra could be the company's last major model release for some time.

Core capabilities

According to OpenAI, Astra performs particularly well at computer-use and browsing tasks, along with software engineering, cybersecurity, science and general professional work. A demonstration video released alongside the launch shows the model handling multiple tasks simultaneously across different domains, such as building 3D models and slideshows.

OpenAI says Astra can manage complex, multi-step workflows both on a computer and within a browser, showing strong visual judgment while staying aligned with its original instructions.

Benchmark results

Astra scored 98.6% on ARC-AGI-3, a benchmark measuring an AI's ability to solve unfamiliar problems, though this score comes with caveats tied to differences in how models are configured. Clearer improvements appeared elsewhere: 57.7% on the Terminal Bench 4.0 coding benchmark and 59.3% on the Agent's Last Exam, both notably higher than GPT-5.6 Sol.

The model's cybersecurity performance stood out in particular. On ExploitBench, which measures a model's ability to exploit software vulnerabilities, Astra achieved a perfect score, compared with 78.5% for GPT-5.6 Sol. On SRE-Bench, which tests reverse-engineering of software without source code, Astra solved 88% of tasks on the first attempt and 99.2% within four attempts.

OpenAI says the model has been built to refuse advanced cybersecurity tasks and includes stronger protections against jailbreak attempts, along with improved monitoring for misuse.

Availability and pricing

GPT-6 Astra is initially rolling out to a limited set of organizations and will become available in the coming days to all ChatGPT Plus, Pro, Business and Enterprise users, as well as through the OpenAI API and AWS. For developers, API access costs $10 per million input tokens and $50 per million output tokens, putting it among the pricier models on the market.

Comments

0/1500

Comments are automatically moderated. No hate, threats, personal data or spam.

Loading comments…

More in this category