OpenAI has unveiled its next-generation artificial intelligence (AI) model, GPT-6 Astra, hailing it as the dawn of artificial general intelligence (AGI).
In a statement released on Thursday, OpenAI introduced GPT-6 Astra, describing it as their most intelligent and aligned model to date.
GPT-6 Astra has set new benchmarks in AI capabilities, scoring an astounding 99.9% on the ARC-AGI-3, an abstract reasoning test designed to evaluate novel problem-solving skills. It also achieved a remarkable 97.6% on the Frontier Math Tier 4 assessment, which challenges AI with complex mathematical problems.
This latest iteration of GPT-6 Astra boasts significantly enhanced agent abilities, allowing it to interact directly with computer systems. In the OSWorld 2.0 test, which measures practical computer usage skills, GPT-6 Astra scored 72.6% in just 40 minutes per task. This marks a substantial improvement over its predecessor, GPT-5.6 Sol, which took 75 minutes to reach 65.7%. Essentially, GPT-6 Astra has achieved higher performance in computer usage while requiring 47% less time.
OpenAI also announced an update to Codex Harness alongside GPT-6 Astra, boasting a work speed 1.9 times faster than its previous version.
The model’s prowess in coding and scientific domains has seen significant advancements. It scored 57.9% on Terminal Bench 4.0 and an impressive 64.6% on Terminal Bench Science 0.1, far outperforming its predecessor’s 22.4% in the scientific realm.
GPT-6 Astra’s capacity for processing extensive contexts has been expanded, now supporting a context window of up to 1.05 million tokens based on application programming interface (API) standards.
The initial rollout of GPT-6 Astra will be limited to select institutions. However, OpenAI plans to make it available soon to ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API and AWS.
During a press briefing, Greg Brockman, President of OpenAI, introduced GPT-6 Astra with the bold statement, Welcome to the era of AGI.