Anthropic Weighs New AI Model Amid Competitive Pressure and Safety Scrutiny

Anthropic is evaluating the launch of a new AI model to defend its enterprise market position against OpenAI, even as the company faces internal and external pressure to prioritize safety over rapid capability expansion. The potential release comes as OpenAI's recently launched GPT-6 Astra gains significant traction among corporate users, challenging Anthropic's status as a leading provider of enterprise AI tools.

OpenAI's release of GPT-6 Astra on September 3 has disrupted the enterprise AI landscape, with the model showing strong performance in software engineering, cybersecurity, and computer use. Data from the corporate expense platform Ramp indicates that Astra has captured approximately 13% of enterprise AI spending, compared to 8% for Anthropic's Claude Fable. Furthermore, OpenRouter, a platform that tracks developer traffic across various models, reported that users spent more on OpenAI models than Anthropic models for the first time in over two and a half years.

Despite these shifts, Anthropic maintains a substantial revenue lead. The company reported an annualized revenue run rate exceeding $65 billion by the end of July, compared to OpenAI's $40 billion. Investors remain divided on whether Astra poses a long-term threat, noting that the enterprise market often favors incumbent vendors due to the complexity of switching infrastructure.

Anthropic is balancing these competitive demands with a commitment to safety, recently announcing a $2 billion partnership with Accenture to fund independent model evaluation. Under this five-year initiative, each company will contribute at least $1 billion to support embedded evaluation, where independent experts work within AI labs to monitor model development and verify safety compliance. This move follows a 3,800-word essay published by Anthropic CEO Dario Amodei on September 12, in which he urged the industry to slow the pace of capability improvements to mitigate risks associated with autonomous AI agents.

Beyond the rivalry with OpenAI, Anthropic faces broader industry challenges, including the rise of open-source and open-weight models that allow companies to build infrastructure without relying on commercial providers. Meta Platforms, previously a major customer, is reportedly developing more internal AI capabilities, potentially reducing its reliance on external models. Meanwhile, Anthropic continues to prepare for an initial public offering, which has been delayed and is now expected to occur after the November US midterm elections.