OpenAI Unveils GPT-6 Astra, Claims It May Mark the Start of the AGI Era

OpenAI announced Thursday the release of its next-generation AI model, GPT-6 Astra, which it describes as top-tier at handling computer and web browser tasks, writing software, and solving complex math problems. The model will roll out to paying customers in the coming days, starting with enterprise clients in the Daybreak early access program, then expanding to ChatGPT Plus, Pro, and Business subscribers. OpenAI did not confirm whether free ChatGPT users will get access.

During a press briefing, OpenAI executives expressed strong confidence in the new model. President and cofounder Greg Brockman said he personally believes the world has entered the era of artificial general intelligence-AI systems generally smarter than humans. "It's not unreasonable to feel that we are now in the AGI era," he told reporters, adding that looking back in a few years, people may pinpoint this time and this model as the true origin of AGI.

The launch comes at a pivotal time for OpenAI, which is pushing to accelerate technology development and boost revenue ahead of a planned initial public offering, while also trying to uphold safety and security standards. The company faces stiff competition from rival Anthropic, which is also preparing its own IPO. In a blog post, OpenAI called GPT-6 Astra the "world's best computer use model," claiming it can navigate browsers and software far more effectively than earlier versions. Historically, AI models have struggled with reliable web browsing, form filling, and tool manipulation. OpenAI says in tests, GPT-6 Astra booked DMV appointments, searched for jobs, and hunted for apartments faster than the average person.

OpenAI also asserts that GPT-6 Astra is its safest and most aligned model yet. Chief scientist Jakub Pachocki told reporters the company is monitoring the model's chain-of-thought-its internal reasoning process-to ensure safe deployment. However, he acknowledged that ensuring advanced AI doesn't cause unintended harm is becoming harder, and this could slow future progress. "We think confidence in monitoring may constrain further development," Pachocki said, noting that OpenAI would withhold scaling if it couldn't maintain sufficient confidence in its ability to monitor model alignment.

Source: Wired