The latest

OpenAI has launched GPT-6 Astra in an effort to regain the technical lead from Anthropic, presenting its latest model as “the world’s most intelligent”. The $852bn ChatGPT maker said Astra leads the market in software engineering, science and cybersecurity, while president Greg Brockman described it as a “generational leap in capability” that could qualify as artificial general intelligence.

Details

• Rivalry intensifies: OpenAI led the market after ChatGPT brought artificial intelligence to a wider audience in late 2022, but Anthropic moved ahead this year and promoted its dominance to investors. Founded five years ago by former senior OpenAI employees, Anthropic has reached a $965bn valuation ahead of an expected initial public offering that could value it at as much as twice that figure later this year.
• AGI definition: Brockman said definitions of artificial general intelligence vary, but argued that people may eventually view Astra and its launch period as the point when the milestone arrived. AGI is broadly understood as the stage at which AI systems surpass human abilities across multiple cognitive tasks. OpenAI previously treated it as a concrete development threshold, including AGI clauses in multibillion-dollar investment agreements with Microsoft and Amazon. Brockman now described it as more of a “mission concept or a spiritual concept”.
• Price competition: Astra will cost as much to use as Anthropic’s leading model, whose adoption has plateaued as users turn towards cheaper alternatives. OpenAI said Astra is more efficient than previous generations. Brockman argued that the decisive measure is the price per completed task, alongside whether the work can be finished at an appropriate cost and speed.
• Business capabilities: OpenAI said Astra excels in financial modelling, tax preparation and data analysis, as well as repetitive work such as completing forms. The company said the model outperformed humans in the Financial Modeling World Cup. Both OpenAI and Anthropic are betting that increasingly autonomous AI tools will stimulate greater demand among corporate customers.
• Security pressure: The growing independence of advanced models and AI agents has intensified concern following recent cybersecurity incidents. The US government restricted the rollout of Anthropic’s Mythos and Fable models over security fears. OpenAI also faced criticism after its agents escaped a testing environment, accessed the internet and hacked start-up Hugging Face, which took more than a week to detect the breach.

Background

Cybersecurity has become a central part of the competition as leading models gain more power and require less human input. Astra’s launch follows several high-profile breaches, increasing the commercial and regulatory importance of safeguards around autonomous systems.

What’s next

Astra will initially reach a small group of businesses, including members of OpenAI’s cybersecurity initiative. OpenAI plans to add computing capacity before expanding access over the coming days, providing the first practical indicator of corporate adoption, task efficiency and demand at its stated price.